Contact Geometry - Mathematisches Institut der Universität zu Köln

Contents
Contact Geometry
Hansjörg Geiges
Mathematisches Institut, Universität zu Köln,
Weyertal 86–90, 50931 Köln, Germany
E-mail: geiges@math.uni-koeln.de
April 2004
1 Introduction 3
2 Contact manifolds 4
2.1 Contact manifolds and their submanifolds . . . . . . . . . . . . . . 6
2.2 Gray stability and the Moser trick . . . . . . . . . . . . . . . . . . 13
2.3 Contact Hamiltonians . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.4 Darboux’s theorem and neighbourhood theorems . . . . . . . . . . 17
2.4.1 Darboux’s theorem . . . . . . . . . . . . . . . . . . . . . . . 17
2.4.2 Isotropic submanifolds . . . . . . . . . . . . . . . . . . . . . 19
2.4.3 Contact submanifolds . . . . . . . . . . . . . . . . . . . . . 24
2.4.4 Hypersurfaces . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.4.5 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.5 Isotopy extension theorems . . . . . . . . . . . . . . . . . . . . . . 32
2.5.1 Isotropic submanifolds . . . . . . . . . . . . . . . . . . . . . 32
2.5.2 Contact submanifolds . . . . . . . . . . . . . . . . . . . . . 34
2.5.3 Surfaces in 3–manifolds . . . . . . . . . . . . . . . . . . . . 36
2.6 Approximation theorems . . . . . . . . . . . . . . . . . . . . . . . . 37
2.6.1 Legendrian knots . . . . . . . . . . . . . . . . . . . . . . . . 38
2.6.2 Transverse knots . . . . . . . . . . . . . . . . . . . . . . . . 42
1

3 Contact structures on 3–manifolds 43
3.1 An invariant of transverse knots . . . . . . . . . . . . . . . . . . . . 45
3.2 Martinet’s construction . . . . . . . . . . . . . . . . . . . . . . . . 46
3.3 2–plane fields on 3–manifolds . . . . . . . . . . . . . . . . . . . . . 50
3.3.1 Hopf’s Umkehrhomomorphismus . . . . . . . . . . . . . . . 53
3.3.2 Representing homology classes by submanifolds . . . . . . . 54
3.3.3 Framed cobordisms . . . . . . . . . . . . . . . . . . . . . . . 56
3.3.4 Definition of the obstruction classes . . . . . . . . . . . . . 57
3.4 Let’s Twist Again . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
3.5 Other existence proofs . . . . . . . . . . . . . . . . . . . . . . . . . 64
3.5.1 Open books . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
3.5.2 Branched covers . . . . . . . . . . . . . . . . . . . . . . . . 66
3.5.3 . . . and more . . . . . . . . . . . . . . . . . . . . . . . . . 67
3.6 Tight and overtwisted . . . . . . . . . . . . . . . . . . . . . . . . . 67
3.7 Classification results . . . . . . . . . . . . . . . . . . . . . . . . . . 73
4 A guide to the literature 75
4.1 Dimension 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
4.2 Higher dimensions . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
4.3 Symplectic fillings . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
4.4 Dynamics of the Reeb vector field . . . . . . . . . . . . . . . . . . . 77
2

1 Introduction
Over the past two decades, contact geometry has undergone a veritable meta-
morphosis: once the ugly duckling known as ‘the odd-dimensional analogue of
symplectic geometry’, it has now evolved into a proud field of study in its own
right. As is typical for a period of rapid development in an area of mathematics,
there are a fair number of folklore results that every mathematician working in
the area knows, but no references that make these results accessible to the novice.
I therefore take the present article as an opportunity to take stock of some of that
folklore.
There are many excellent surveys covering specific aspects of contact geometry
(e.g. classification questions in dimension 3, dynamics of the Reeb vector field,
various notions of symplectic fillability, transverse and Legendrian knots and
links). All these topics deserve to be included in a comprehensive survey, but
an attempt to do so here would have left this article in the ‘to appear’ limbo for
much too long.
Thus, instead of adding yet another survey, my plan here is to cover in detail
some of the more fundamental differential topological aspects of contact geometry.
In doing so, I have not tried to hide my own idiosyncrasies and preoccupations.
Owing to a relatively leisurely pace and constraints of the present format, I
have not been able to cover quite as much material as I should have wished.
Nonetheless, I hope that the reader of the present handbook chapter will be
better prepared to study some of the surveys I alluded to – a guide to these
surveys will be provided – and from there to move on to the original literature.
A book chapter with comparable aims is Chapter 8 in [1]. It seemed opportune
to be brief on topics that are covered extensively there, even if it is done at the
cost of leaving out some essential issues. I hope to return to the material of the
present chapter in a yet to be written more comprehensive monograph.
Acknowledgements. I am grateful to Fan Ding, Jesús Gonzalo and Federica
Pasquotto for their attentive reading of the original manuscript. I also thank
John Etnyre and Stephan Schönenberger for allowing me to use a couple of their
figures (viz., Figures 2 and 1 of the present text, respectively).
3

2 Contact manifolds
Let M be a differential manifold and ξ ⊂ TM a field of hyperplanes on M. Locally
such a hyperplane field can always be written as the kernel of a non-vanishing
1–form α. One way to see this is to choose an auxiliary Riemannian metric g on
M and then to define α = g(X, .), where X is a local non-zero section of the line
bundle ξ ⊥ (the orthogonal complement of ξ in TM). We see that the existence
of a globally defined 1–form α with ξ = kerα is equivalent to the orientability
(hence triviality) of ξ ⊥ , i.e. the coorientability of ξ. Except for an example below,
I shall always assume this condition.
If α satisfies the Frobenius integrability condition
α ∧ dα = 0,
then ξ is an integrable hyperplane field (and vice versa), and its integral sub-
manifolds form a codimension 1 foliation of M. Equivalently, this integrability
condition can be written as
X, Y ∈ ξ =⇒ [X, Y ] ∈ ξ.
An integrable hyperplane field is locally of the form dz = 0, where z is a coordi-
nate function on M. Much is known, too, about the global topology of foliations,
cf. [100].
Contact structures are in a certain sense the exact opposite of integrable
hyperplane fields.
Definition 2.1. Let M be a manifold of odd dimension 2n + 1. A contact
structure is a maximally non-integrable hyperplane field ξ = kerα ⊂ TM, that
is, the defining 1–form α is required to satisfy
α ∧ (dα) n �= 0
(meaning that it vanishes nowhere). Such a 1–form α is called a contact form.
The pair (M, ξ) is called a contact manifold.
Remark 2.2. Observe that in this case α ∧ (dα) n is a volume form on M; in
particular, M needs to be orientable. The condition α∧(dα) n �= 0 is independent
of the specific choice of α and thus is indeed a property of ξ = kerα: Any other 1–
form defining the same hyperplane field must be of the form λα for some smooth
4

function λ: M → R \ {0}, and we have
(λα) ∧ (d(λα)) n = λα ∧ (λ dα + dλ ∧ α) n = λ n+1 α ∧ (dα) n �= 0.
We see that if n is odd, the sign of this volume form depends only on ξ, not
the choice of α. This makes it possible, given an orientation of M, to speak of
positive and negative contact structures.
Remark 2.3. An equivalent formulation of the contact condition is that we
have (dα) n |ξ �= 0. In particular, for every point p ∈ M, the 2n–dimensional
subspace ξp ⊂ TpM is a vector space on which dα defines a skew-symmetric form
of maximal rank, that is, (ξp, dα|ξp ) is a symplectic vector space. A consequence
of this fact is that there exists a complex bundle structure J : ξ → ξ compatible
with dα (see [92, Prop. 2.63]), i.e. a bundle endomorphism satisfying
• J 2 = −idξ,
• dα(JX, JY ) = dα(X, Y ) for all X, Y ∈ ξ,
• dα(X, JX) > 0 for 0 �= X ∈ ξ.
Remark 2.4. The name ‘contact structure’ has its origins in the fact that one of
the first historical sources of contact manifolds are the so-called spaces of contact
elements (which in fact have to do with ‘contact’ in the differential geometric
sense), see [7] and [45].
In the 3–dimensional case the contact condition can also be formulated as
X, Y ∈ ξ linearly independent =⇒ [X, Y ] �∈ ξ;
this follows immediately from the equation
dα(X, Y ) = X(α(Y )) − Y (α(X)) − α([X, Y ])
and the fact that the contact condition (in dim. 3) may be written as dα|ξ �= 0.
In the present article I shall take it for granted that contact structures are
worthwhile objects of study. As I hope to illustrate, this is fully justified by
the beautiful mathematics to which they have given rise. For an apology of
contact structures in terms of their origin (with hindsight) in physics and the
multifarious connections with other areas of mathematics I refer the reader to the
5

historical surveys [87] and [45]. Contact structures may also be justified on the
grounds that they are generic objects: A generic 1–form α on an odd-dimensional
manifold satisfies the contact condition outside a smooth hypersurface, see [89].
Similarly, a generic 1–form α on a 2n–dimensional manifold satisfies the condition
α ∧ (dα) n−1 �= 0 outside a submanifold of codimension 3; such ‘even-contact
manifolds’ have been studied in [51], for instance, but on the whole their theory
is not as rich or well-motivated as that of contact structures.
Definition 2.5. Associated with a contact form α one has the so-called Reeb
vector field Rα, defined by the equations
(i) dα(Rα, .) ≡ 0,
(ii) α(Rα) ≡ 1.
As a skew-symmetric form of maximal rank 2n, the form dα|TpM has a 1–
dimensional kernel for each p ∈ M 2n+1 . Hence equation (i) defines a unique
line field 〈Rα〉 on M. The contact condition α ∧ (dα) n �= 0 implies that α is
non-trivial on that line field, so a global vector field is defined by the additional
normalisation condition (ii).
2.1 Contact manifolds and their submanifolds
We begin with some examples of contact manifolds; the simple verification that
the listed 1–forms are contact forms is left to the reader.
Example 2.6. On R 2n+1 with cartesian coordinates (x1, y1, . . .,xn, yn, z), the
1–form
is a contact form.
α1 = dz +
n�
j=1
xj dyj
Example 2.7. On R 2n+1 with polar coordinates (rj, ϕj) for the (xj, yj)–plane,
j = 1, . . .,n, the 1–form
is a contact form.
α2 = dz +
n�
j=1
r 2 j dϕj = dz +
6
n�
(xj dyj − yj dxj)
j=1

x
z
Figure 1: The contact structure ker(dz + x dy).
Definition 2.8. Two contact manifolds (M1, ξ1) and (M2, ξ2) are called contac-
tomorphic if there is a diffeomorphism f : M1 → M2 with Tf(ξ1) = ξ2, where
Tf : TM1 → TM2 denotes the differential of f. If ξi = kerαi, i = 1, 2, this
is equivalent to the existence of a nowhere zero function λ: M1 → R such that
f ∗ α2 = λα1.
Example 2.9. The contact manifolds (R 2n+1 , ξi = kerαi), i = 1, 2, from the
preceding examples are contactomorphic. An explicit contactomorphism f with
f ∗ α2 = α1 is given by
f(x, y, z) = � (x + y)/2, (y − x)/2, z + xy/2 � ,
where x and y stand for (x1, . . .,xn) and (y1, . . .,yn), respectively, and xy stands
for �
j xjyj. Similarly, both these contact structures are contactomorphic to
ker(dz − �
j yj dxj). Any of these contact structures is called the standard
contact structure on R 2n+1 .
Example 2.10. The standard contact structure on the unit sphere S 2n+1
in R 2n+2 (with cartesian coordinates (x1, y1, . . .,xn+1, yn+1)) is defined by the
contact form
n+1 �
α0 = (xj dyj − yj dxj).
j=1
With r denoting the radial coordinate on R2n+2 (that is, r2 = �
j (x2j + y2 j )) one
checks easily that α0 ∧ (dα0) n ∧ r dr �= 0 for r �= 0. Since S2n+1 is a level surface
of r (or r 2 ), this verifies the contact condition.
7
y

Alternatively, one may regard S 2n+1 as the unit sphere in C n+1 with complex
structure J (corresponding to complex coordinates zj = xj+iyj, j = 1, . . .,n+1).
Then ξ0 = kerα0 defines at each point p ∈ S 2n+1 the complex (i.e. J–invariant)
subspace of TpS 2n+1 , that is,
ξ0 = TS 2n+1 ∩ J(TS 2n+1 ).
This follows from the observation that α = −r dr◦J. The hermitian form dα(., J.)
on ξ0 is called the Levi form of the hypersurface S 2n+1 ⊂ C n+1 . The contact
condition for ξ corresponds to the positive definiteness of that Levi form, or what
in complex analysis is called the strict pseudoconvexity of the hypersurface. For
more on the question of pseudoconvexity from the contact geometric viewpoint
see [1, Section 8.2]. Beware that the ‘complex structure’ in their Proposition 8.14
is not required to be integrable, i.e. constitutes what is more commonly referred
to as an ‘almost complex structure’.
Definition 2.11. Let (V, ω) be a symplectic manifold of dimension 2n + 2,
that is, ω is a closed (dω = 0) and non-degenerate (ω n+1 �= 0) 2–form on V . A
vector field X is called a Liouville vector field if LXω = ω, where L denotes
the Lie derivative.
With the help of Cartan’s formula LX = d ◦ iX + iX ◦ d this may be rewrit-
ten as d(iXω) = ω. Then the 1–form α = iXω defines a contact form on any
hypersurface M in V transverse to X. Indeed,
α ∧ (dα) n = iXω ∧ (d(iXω)) n = iXω ∧ ω n = 1
n + 1 iX(ω n+1 ),
which is a volume form on M ⊂ V provided M is transverse to X.
Example 2.12. With V = R 2n+2 , symplectic form ω = �
j dxj ∧ dyj, and
Liouville vector field X = �
j (xj∂xj + yj∂yj )/2 = r∂r/2, we recover the standard
contact structure on S 2n+1 .
For finer issues relating to hypersurfaces in symplectic manifolds transverse
to a Liouville vector field I refer the reader to [1, Section 8.2].
Here is a further useful example of contactomorphic manifolds.
Proposition 2.13. For any point p ∈ S 2n+1 , the manifold (S 2n+1 \ {p}, ξ0) is
contactomorphic to (R 2n+1 , ξ2).
8

Proof. The contact manifold (S 2n+1 , ξ0) is a homogeneous space under the nat-
ural U(n + 1)–action, so we are free to choose p = (0, . . .,0, −1). Stereographic
projection from p does almost, but not quite yield the desired contactomorphism.
Instead, we use a map that is well-known in the theory of Siegel domains (cf. [3,
Chapter 8]) and that looks a bit like a complex analogue of stereographic projec-
tion; this was suggested in [92, Exercise 3.64].
Regard S 2n+1 as the unit sphere in C n+1 = C n ×C with cartesian coordinates
(z1, . . .,zn, w) = (z, w). We identify R 2n+1 with C n ×R ⊂ C n ×C with coordinates
(ζ1, . . .,ζn, s) = (ζ, s) = (ζ,Re σ), where ζj = xj + iyj. Then
n�
α2 = ds + (xj dyj − yj dxj)
and
j=1
= ds + i
(ζ dζ − ζ dζ).
2
α0 = i
(z dz − z dz + w dw − w dw).
2
Now define a smooth map f : S 2n+1 \ {(0, −1)} → R 2n+1 by
Then
and
(ζ, s) = f(z, w) =
�
z i(w − w)
, −
1 + w 2|1 + w| 2
�
.
f ∗ i dw i dw
ds = − +
2|1 + w| 2 2|1 + w| 2
+ i(w − w) dw i(w − w) dw
+
2(1 + w) |1 + w| 2 2(1 + w) |1 + w| 2
i
=
2|1 + w| 2
�
w − w w − w
−dw + dw + dw +
1 + w 1 + w dw
�
f ∗ (ζ dζ − ζ dζ) =
Along S 2n+1 we have
=
�
z dz
1 + w
− z
�
dz
z
(1 + w) 2dw
�
1 + w −
1 + w 1 + w −
z
(1 + w) 2dw
�
1
|1 + w| 2
�
z dz − zdz + |z| 2
� ��
dw dw
− .
1 + w 1 + w
|z| 2 = 1 − |w| 2 = (1 − w)(1 + w) + (w − w)
= (1 − w)(1 + w) − (w − w),
9

whence
|z| 2
� �
dw dw
−
1 + w 1 + w
w − w
= (1 − w)dw −
1 + w dw
w − w
− (1 − w)dw −
1 + w dw.
From these calculations we conclude f ∗ α2 = α0/|1 + w| 2 . So it only remains to
show that f is actually a diffeomorphism of S 2n+1 \ {(0, −1)} onto R 2n+1 . To
that end, consider the map
defined by
�f : (C n × C) \ (C n × {−1}) −→ (C n × C) \ (C n × {−i/2})
(ζ, σ) = � f(z, w) =
� �
z i w − 1
, − .
1 + w 2 w + 1
This is a biholomorphic map with inverse map
� �
2ζ 1 + 2iσ
(ζ, σ) ↦−→ , .
1 − 2iσ 1 − 2iσ
We compute
w − 1 w − 1
Im σ = − −
4(w + 1) 4(w + 1)
(w − 1)(w + 1) + (w − 1)(w + 1)
= −
4|1 + w| 2
= 1 − |w|2
2|1 + w| 2.
Hence for (z, w) ∈ S 2n+1 \ {(0, −1)} we have
Imσ =
|z| 2 1
=
2|1 + w| 2 2 |ζ|2 ;
conversely, any point (ζ, σ) with Imσ = |ζ| 2 /2 lies in the image of � f| S 2n+1 \{(0,−1)},
that is, � f restricted to S 2n+1 \{(0, −1)} is a diffeomorphism onto {Im σ = |ζ| 2 /2}.
Finally, we compute
i(w − 1) i(w − 1)
Re σ = − +
4(w + 1) 4(w + 1)
(w − 1)(w + 1) − (w − 1)(w + 1)
= −i
4|1 + w| 2
i(w − w)
= −
2|1 + w| 2,
10

from which we see that for (z, w) ∈ S 2n+1 \ {(0, −1)} and with (ζ, σ) = � f(z, w)
we have f(z, w) = (ζ,Re σ). This concludes the proof.
At the beginning of this section I mentioned that one may allow contact
structures that are not coorientable, and hence not defined by a global contact
form.
Example 2.14. Let M = R n+1 ×RP n with cartesian coordinates (x0, . . .,xn) on
the R n+1 –factor and homogeneous coordinates [y0 : . . . : yn] on the RP n –factor.
Then
ξ = ker � n �
j=0
�
yj dxj
is a well-defined hyperplane field on M, because the 1–form on the right-hand side
is well-defined up to scaling by a non-zero real constant. On the open submanifold
Uk = {yk �= 0} ∼ = Rn+1 × Rn of M we have ξ = kerαk with
αk = dxk + �
� �
yj
dxj
j�=k
an honest 1–form on Uk. This is the standard contact form of Example 2.6, which
proves that ξ is a contact structure on M.
If n is even, then M is not orientable, so there can be no global contact
form defining ξ (cf. Remark 2.2), i.e. ξ is not coorientable. Notice, however, that
a contact structure on a manifold of dimension 2n + 1 with n even is always
orientable: the sign of (dα) n |ξ does not depend on the choice of local 1–form
defining ξ.
If n is odd, then M is orientable, so it would be possible that ξ is the kernel
of a globally defined 1–form. However, since the sign of α ∧ (dα) n , for n odd, is
independent of the choice of local 1–form defining ξ, it is also conceivable that no
global contact form exists. (In fact, this consideration shows that any manifold
of dimension 2n + 1, with n odd, admitting a contact structure (coorientable or
not) needs to be orientable.) This is indeed what happens, as we shall prove now.
Proposition 2.15. Let (M, ξ) be the contact manifold of the preceding example.
Then TM/ξ can be identified with the canonical line bundle on RP n (pulled back
to M). In particular, TM/ξ is a non-trivial line bundle, so ξ is not coorientable.
11
yk

Proof. For given y = [y0 : . . . : yn] ∈ RP n , the vector y0∂x0 +· · ·+yn∂xn ∈ TxR n+1
is well-defined up to a non-zero real factor (and independent of x ∈ R n+1 ), and
hence defines a line ℓy in TxR n+1 ∼ = R n+1 . The set
E = {(t, x, y): x ∈ R n+1 , y ∈ RP n , t ∈ ℓy}
⊂ TR n+1 × RP n ⊂ T(R n+1 × RP n ) = TM
with projection (t, x, y) ↦→ (x, y) defines a line sub-bundle of TM that restricts
to the canonical line bundle over {x} × RP n ≡ RP n for each x ∈ R n+1 . The
canonical line bundle over RP n is well-known to be non-trivial [95, p. 16], so the
same holds for E.
Moreover, E is clearly complementary to ξ, i.e. TM/ξ ∼ = E, since
n�
j=0
yj dxj(
n�
k=0
yk∂xk ) =
This proves that that ξ is not coorientable.
n�
j=0
y 2 j �= 0.
To sum up, in the example above we have one of the following two situations:
• If n is odd, then M is orientable; ξ is neither orientable nor coorientable.
• If n is even, then M is not orientable; ξ is not coorientable, but it is ori-
entable.
We close this section with the definition of the most important types of sub-
manifolds.
Definition 2.16. Let (M, ξ) be a contact manifold.
(i) A submanifold L of (M, ξ) is called an isotropic submanifold if TxL ⊂ ξx
for all x ∈ L.
(ii) A submanifold M ′ of M with contact structure ξ ′ is called a contact
submanifold if TM ′ ∩ ξ|M ′ = ξ′ .
Observe that if ξ = kerα and i: M ′ → M denotes the inclusion map, then the
condition for (M ′ , ξ ′ ) to be a contact submanifold of (M, ξ) is that ξ ′ = ker(i ∗ α).
In particular, ξ ′ ⊂ ξ|M ′ is a symplectic sub-bundle with respect to the symplectic
bundle structure on ξ given by dα.
The following is a manifestation of the maximal non-integrability of contact
structures.
12

Proposition 2.17. Let (M, ξ) be a contact manifold of dimension 2n + 1 and L
an isotropic submanifold. Then dim L ≤ n.
Proof. Write i for the inclusion of L in M and let α be an (at least locally
defined) contact form defining ξ. Then the condition for L to be isotropic becomes
i ∗ α ≡ 0. It follows that i ∗ dα ≡ 0. In particular, TpL ⊂ ξp is an isotropic
subspace of the symplectic vector space (ξp, dα|ξp ), i.e. a subspace on which the
symplectic form restricts to zero. From Linear Algebra we know that this implies
dimTpL ≤ (dimξp)/2 = n.
Definition 2.18. An isotropic submanifold L ⊂ (M 2n+1 , ξ) of maximal possible
dimension n is called a Legendrian submanifold.
In particular, in a 3–dimensional contact manifold there are two distinguished
types of knots: Legendrian knots on the one hand, transverse 1 knots on the
other, i.e. knots that are everywhere transverse to the contact structure. If ξ
is cooriented by a contact form α and γ : S 1 → (M, ξ = kerα) is oriented, one
can speak of a positively or negatively transverse knot, depending on whether
α(˙γ) > 0 or α(˙γ) < 0.
2.2 Gray stability and the Moser trick
The Gray stability theorem that we are going to prove in this section says that
there are no non-trivial deformations of contact structures on closed manifolds.
In fancy language, this means that contact structures on closed manifolds have
discrete moduli. First a preparatory lemma.
Lemma 2.19. Let ωt, t ∈ [0, 1], be a smooth family of differential k–forms on a
manifold M and (ψt) t∈[0,1] an isotopy of M. Define a time-dependent vector field
Xt on M by Xt ◦ ψt = ˙ ψt, where the dot denotes derivative with respect to t (so
that ψt is the flow of Xt). Then
d � � � � ∗ ∗
ψt ωt = ψt ˙ωt + LXtωt .
dt
Proof. For a time-independent k–form ω we have
d � ∗
ψt ω
dt
� = ψ ∗� t LXtω � .
This follows by observing that
1 Some people like to call them ‘transversal knots’, but I adhere to J.H.C. Whitehead’s dictum,
as quoted in [64]: “Transversal is a noun; the adjective is transverse.”
13

(i) the formula holds for functions,
(ii) if it holds for differential forms ω and ω ′ , then also for ω ∧ ω ′ ,
(iii) if it holds for ω, then also for dω,
(iv) locally functions and differentials of functions generate the algebra of dif-
ferential forms.
We then compute
d
dt (ψ∗ ψ
t ωt) = lim
h→0
∗ t+hωt+h − ψ∗ t ωt
h
= lim
h→0
ψ ∗ t+h ωt+h − ψ ∗ t+h ωt + ψ ∗ t+h ωt − ψ ∗ t ωt
= lim ψ
h→0 ∗ � �
ωt+h − ωt
t+h + lim
h h→0
� �
˙ωt + LXtωt .
= ψ ∗ t
h
ψ ∗ t+h ωt − ψ ∗ t ωt
h
For that last equality observe (regarding the second summand) that ψt+h =
ψt h ◦ ψt, where ψt h
dependent vector field Xt h := Xt+h; then apply the result for time-independent
k–forms.
denotes, for fixed t and time-variable h, the flow of the time-
Theorem 2.20 (Gray stability). Let ξt, t ∈ [0, 1], be a smooth family of contact
structures on a closed manifold M. Then there is an isotopy (ψt) t∈[0,1] of M such
that
Tψt(ξ0) = ξt for each t ∈ [0, 1].
Proof. The simplest proof of this result rests on what is known as the Moser
trick, introduced by J. Moser [96] in the context of stability results for (equicoho-
mologous) volume and symplectic forms. J. Gray’s original proof [61] was based
on deformation theory à la Kodaira-Spencer. The idea of the Moser trick is to
assume that ψt is the flow of a time-dependent vector field Xt. The desired equa-
tion for ψt then translates into an equation for Xt. If that equation can be solved,
the isotopy ψt is found by integrating Xt; on a closed manifold the flow of Xt will
be globally defined.
Let αt be a smooth family of 1–forms with kerαt = ξt. The equation in the
theorem then translates into
ψ ∗ t αt = λtα0,
14

where λt: M → R + is a suitable smooth family of smooth functions. Differen-
tiation of this equation with respect to t yields, with the help of the preceding
lemma,
ψ ∗� �
t ˙αt + LXtαt = λtα0
˙ = ˙ λt
ψ
λt
∗ t αt,
or, with the help of Cartan’s formula LX = d ◦ iX + iX ◦ d and with µt =
d
dt (log λt) ◦ ψ −1
t ,
ψ ∗� � ∗
t ˙αt + d(αt(Xt)) + iXtdαt = ψt (µtαt).
If we choose Xt ∈ ξt, this equation will be satisfied if
Plugging in the Reeb vector field Rαt gives
˙αt + iXtdαt = µtαt. (2.1)
˙αt(Rαt) = µt. (2.2)
So we can use (2.2) to define µt, and then the non-degeneracy of dαt|ξt and
the fact that Rαt ∈ ker(µtαt − ˙αt) allow us to find a unique solution Xt ∈ ξt
of (2.1).
Remark 2.21. (1) Contact forms do not satisfy stability, that is, in general
one cannot find an isotopy ψt such that ψ ∗ t αt = α0. For instance, consider the
following family of contact forms on S 3 ⊂ R 4 :
αt = (x1 dy1 − y1 dx1) + (1 + t)(x2 dy2 − y2 dx2),
where t ≥ 0 is a real parameter. The Reeb vector field of αt is
Rαt = (x1 ∂y1 − y1
1
∂x1 ) +
1 + t (x2 ∂y2 − y2 ∂x2 ).
The flow of Rα0 defines the Hopf fibration, in particular all orbits of Rα0 are
closed. For t ∈ R + \ Q, on the other hand, Rαt has only two periodic orbits. So
there can be no isotopy with ψ ∗ t αt = α0, because such a ψt would also map Rα0
to Rαt.
(2) Y. Eliashberg [25] has shown that on the open manifold R 3 there are
likewise no non-trivial deformations of contact structures, but on S 1 × R 2 there
does exist a continuum of non-equivalent contact structures.
(3) For further applications of this theorem it is useful to observe that at
points p ∈ M with ˙αt,p identically zero in t we have Xt(p) ≡ 0, so such points
remain stationary under the isotopy ψt.
15

2.3 Contact Hamiltonians
A vector field X on the contact manifold (M, ξ = kerα) is called an infinitesi-
mal automorphism of the contact structure if the local flow of X preserves ξ
(The study of such automorphisms was initiated by P. Libermann, cf. [80]). By
slight abuse of notation, we denote this flow by ψt; if M is not closed, ψt (for a
fixed t �= 0) will not in general be defined on all of M. The condition for X to
be an infinitesimal automorphism can be written as Tψt(ξ) = ξ, which is equiv-
alent to LXα = λα for some function λ: M → R (notice that this condition is
independent of the choice of 1–form α defining ξ). The local flow of X preserves
α if and only if LXα = 0.
Theorem 2.22. With a fixed choice of contact form α there is a one-to-one
correspondence between infinitesimal automorphisms X of ξ = kerα and smooth
functions H : M → R. The correspondence is given by
• X ↦−→ HX = α(X);
• H ↦−→ XH, defined uniqely by α(XH) = H and iXH dα = dH(Rα)α − dH.
The fact that XH is uniquely defined by the equations in the theorem follows
as in the preceding section from the fact that dα is non-degenerate on ξ and
Rα ∈ ker(dH(Rα)α − dH).
Proof. Let X be an infinitesimal automorphism of ξ. Set HX = α(X) and write
dHX + iXdα = LXα = λα with λ: M → R. Applying this last equation to
Rα yields dHX(Rα) = λ. So X satisfies the equations α(X) = HX and iXdα =
dHX(Rα)α − dHX. This means that XHX = X.
Conversely, given H : M → R and with XH as defined in the theorem, we
have
LXH α = iXH dα + d(α(XH)) = dH(Rα)α,
so XH is an infinitesimal automorphism of ξ. Moreover, it is immediate from the
definitions that HXH = α(XH) = H.
Corollary 2.23. Let (M, ξ = kerα) be a closed contact manifold and Ht: M →
R, t ∈ [0, 1], a smooth family of functions. Let Xt = XHt be the correspond-
ing family of infinitesimal automorphisms of ξ (defined via the correspondence
described in the preceding theorem). Then the globally defined flow ψt of the
16

time-dependent vector field Xt is a contact isotopy of (M, ξ), that is, ψ ∗ t α = λtα
for some smooth family of functions λt: M → R + .
Proof. With Lemma 2.19 and the preceding proof we have
d � ∗
ψt α
dt
� = ψ ∗� t LXtα � = ψ ∗� t dHt(Rα)α � = µtψ ∗ t α
with µt = dHt(Rα) ◦ ψt. Since ψ0 = idM (whence ψ∗ 0α = α) this implies that,
with
we have ψ ∗ t α = λtα.
λt = exp �� t
µs ds � ,
0
This corollary will be used in Section 2.5 to prove various isotopy extension
theorems from isotopies of special submanifolds to isotopies of the ambient con-
tact manifold. In a similar vein, contact Hamiltonians can be used to show that
standard general position arguments from differential topology continue to hold
in the contact geometric setting. Another application of contact Hamiltonians
is a proof of the fact that the contactomorphism group of a connected contact
manifold acts transitively on that manifold [12]. (See [8] for more on the general
structure of contactomorphism groups.)
2.4 Darboux’s theorem and neighbourhood theorems
The flexibility of contact structures inherent in the Gray stability theorem and
the possibility to construct contact isotopies via contact Hamiltonians results in
a variety of theorems that can be summed up as saying that there are no local
invariants in contact geometry. Such theorems form the theme of the present
section.
In contrast with Riemannian geometry, for instance, where the local structure
coming from the curvature gives rise to a rich theory, the interesting questions
in contact geometry thus appear only at the global level. However, it is actually
that local flexibility that allows us to prove strong global theorems, such as the
existence of contact structures on certain closed manifolds.
2.4.1 Darboux’s theorem
Theorem 2.24 (Darboux’s theorem). Let α be a contact form on the (2n +
1)–dimensional manifold M and p a point on M. Then there are coordinates
17

x1, . . .,xn, y1, . . .,yn, z on a neighbourhood U ⊂ M of p such that
α|U = dz +
n�
j=1
xj dyj.
Proof. We may assume without loss of generality that M = R 2n+1 and p = 0 is
the origin of R 2n+1 . Choose linear coordinates x1, . . .,xn, y1, . . .yn, z on R 2n+1
such that
on T0R 2n+1 :
�
α(∂z) = 1, i∂zdα = 0,
∂xj , ∂yj ∈ kerα (j = 1, . . .,n), dα = � n
j=1 dxj ∧ dyj.
This is simply a matter of linear algebra (the normal form theorem for skew-
symmetric forms on a vector space).
Now set α0 = dz + �
j xj dyj and consider the family of 1–forms
αt = (1 − t)α0 + tα, t ∈ [0, 1],
on R 2n+1 . Our choice of coordinates ensures that
αt = α, dαt = dα at the origin.
Hence, on a sufficiently small neighbourhood of the origin, αt is a contact form
for all t ∈ [0, 1].
We now want to use the Moser trick to find an isotopy ψt of a neighbourhood
of the origin such that ψ ∗ t αt = α0. This aim seems to be in conflict with our
earlier remark that contact forms are not stable, but as we shall see presently,
locally this equation can always be solved.
Indeed, differentiating ψ ∗ t αt = α0 (and assuming that ψt is the flow of some
time-dependent vector field Xt) we find
so Xt needs to satisfy
ψ ∗� �
t ˙αt + LXtαt = 0,
˙αt + d(αt(Xt)) + iXtdαt = 0. (2.3)
Write Xt = HtRαt + Yt with Yt ∈ ker αt. Inserting Rαt in (2.3) gives
˙αt(Rαt) + dHt(Rαt) = 0. (2.4)
18

On a neighbourhood of the origin, a smooth family of functions Ht satisfying
(2.4) can always be found by integration, provided only that this neighbourhood
has been chosen so small that none of the Rαt has any closed orbits there. Since
αt ˙ is zero at the origin, we may require that Ht(0) = 0 and dHt|0 = 0 for all
t ∈ [0, 1]. Once Ht has been chosen, Yt is defined uniquely by (2.3), i.e. by
˙αt + dHt + iYtdαt = 0.
Notice that with our assumptions on Ht we have Xt(0) = 0 for all t.
Now define ψt to be the local flow of Xt. This local flow fixes the origin, so
there it is defined for all t ∈ [0, 1]. Since the domain of definition in R × M of a
local flow on a manifold M is always open (cf. [15, 8.11]), we can infer 2 that ψt
is actually defined for all t ∈ [0, 1] on a sufficiently small neighbourhood of the
origin in R 2n+1 . This concludes the proof of the theorem (strictly speaking, the
local coordinates in the statement of the theorem are the coordinates xj ◦ ψ −1
1
etc.).
Remark 2.25. The proof of this result given in [1] is incomplete: It is not
possible, as is suggested there, to prove the Darboux theorem for contact forms
if one requires Xt ∈ ker αt.
2.4.2 Isotropic submanifolds
Let L ⊂ (M, ξ = kerα) be an isotropic submanifold in a contact manifold with
cooriented contact structure. Write (TL) ⊥ ⊂ ξ|L for the sub-bundle of ξ|L that is
symplectically orthogonal to TL with respect to the symplectic bundle structure
dα|ξ. The conformal class of this symplectic bundle structure depends only on
the contact structure ξ, not on the choice of contact form α defining ξ: If α is
replaced by λα for some smooth function λ: M → R + , then d(λα)|ξ = λ dα|ξ.
So the bundle (TL) ⊥ is determined by ξ.
The fact that L is isotropic implies TL ⊂ (TL) ⊥ . Following Weinstein [105],
we call the quotient bundle (TL) ⊥ /TL with the conformal symplectic structure
induced by dα the conformal symplectic normal bundle of L in M and write
CSN(M, L) = (TL) ⊥ /TL.
2 To be absolutely precise, one ought to work with a family αt, t ∈ R, where αt ≡ α0 for
t ≤ ε and αt ≡ α1 for t ≥ 1 − ε, i.e. a technical homotopy in the sense of [15]. Then Xt will be
defined for all t ∈ R, and the reasoning of [15] can be applied.
19

So the normal bundle NL = (TM|L)/TL of L in M can be split as
NL ∼ = (TM|L)/(ξ|L) ⊕ (ξ|L)/(TL) ⊥ ⊕ CSN(M, L).
Observe that if dimM = 2n + 1 and dimL = k ≤ n, then the ranks of the
three summands in this splitting are 1, k and 2(n − k), respectively. Our aim
in this section is to show that a neighbourhood of L in M is determined, up to
contactomorphism, by the isomorphism type (as a conformal symplectic bundle)
of CSN(M, L).
The bundle (TM|L)/(ξ|L) is a trivial line bundle because ξ is cooriented.
The bundle (ξ|L)/(TL) ⊥ can be identified with the cotangent bundle T ∗ L via the
well-defined bundle isomorphism
Ψ: (ξ|L)/(TL) ⊥ −→ T ∗ L
Y ↦−→ iY dα|TL.
(Ψ is obviously injective and well-defined by the definition of (TL) ⊥ , and the
ranks of the two bundles are equal.)
Although Ψ is well-defined on the quotient (ξ|L)/(TL) ⊥ , to proceed further
we need to choose an isotropic complement of (TL) ⊥ in ξ|L. Restricted to each
fibre ξp, p ∈ L, such an isotropic complement of (TpL) ⊥ exists. There are two
ways to obtain a smooth bundle of such isotropic complements. The first would
be to carry over Arnold’s corresponding discussion of Lagrangian subbundles
of symplectic bundles [6] to the isotropic case in order to show that the space
of isotropic complements of U ⊥ ⊂ V , where U is an isotropic subspace in a
symplectic vector space V , is convex. (This argument uses generating functions
for isotropic subspaces.) Then by a partition of unity argument the desired
complement can be constructed on the bundle level.
A slightly more pedestrian approach is to define this isotropic complement
with the help of a complex bundle structure J on ξ compatible with dα (cf.
Remark 2.3). The condition dα(X, JX) > 0 for 0 �= X ∈ ξ implies that (TpL) ⊥ ∩
J(TpL) = {0} for all p ∈ L, and so a dimension count shows that J(TL) is indeed
a complement of (TL) ⊥ in ξ|L. (In a similar vein, CSN(M, L) can be identified
as a sub-bundle of ξ, viz., the orthogonal complement of TL ⊕ J(TL) ⊂ ξ with
respect to the bundle metric dα(., J.) on ξ.)
On the Whitney sum TL ⊕ T ∗ L (for any manifold L) there is a canonical
symplectic bundle structure ΩL defined by
ΩL,p(X + η, X ′ + η ′ ) = η(X ′ ) − η ′ (X) for X, X ′ ∈ TpL; η, η ′ ∈ T ∗ p L.
20

Lemma 2.26. The bundle map
idTL ⊕ Ψ: (TL ⊕ J(TL), dα) −→ (TL ⊕ T ∗ L,ΩL)
is an isomorphism of symplectic vector bundles.
Proof. We only need to check that idTL ⊕ Ψ is a symplectic bundle map. Let
X, X ′ ∈ TpL and Y, Y ′ ∈ Jp(TpL). Write Y = JpZ, Y ′ = JpZ ′ with Z, Z ′ ∈ TpL.
It follows that
dα(Y, Y ′ ) = dα(JZ, JZ ′ ) = dα(Z, Z ′ ) = 0,
since L is an isotropic submanifold. For the same reason dα(X, X ′ ) = 0. Hence
dα(X + Y, X ′ + Y ′ ) = dα(Y, X ′ ) − dα(Y ′ , X)
= Ψ(Y )(X ′ ) − Ψ(Y ′ )(X)
= ΩL(X + Ψ(Y ), X ′ + Ψ(Y ′ )).
Theorem 2.27. Let (Mi, ξi), i = 0, 1, be contact manifolds with closed isotropic
submanifolds Li. Suppose there is an isomorphism of conformal symplectic nor-
mal bundles Φ: CSN(M0, L0) → CSN(M1, L1) that covers a diffeomorphism
φ: L0 → L1. Then φ extends to a contactomorphism ψ: N(L0) → N(L1) of
suitable neighbourhoods N(Li) of Li such that Tψ| CSN(M0,L0) and Φ are bundle
homotopic (as symplectic bundle isomorphisms).
Corollary 2.28. Diffeomorphic (closed) Legendrian submanifolds have contac-
tomorphic neighbourhoods.
Proof. If Li ⊂ Mi is Legendrian, then CSN(Mi, Li) has rank 0, so the conditions
in the theorem, apart from the existence of a diffeomorphism φ: L0 → L1, are
void.
Example 2.29. Let S 1 ⊂ (M 3 , ξ) be a Legendrian knot in a contact 3–manifold.
Then with a coordinate θ ∈ [0, 2π] along S 1 and coordinates x, y in slices trans-
verse to S 1 , the contact structure
cos θ dx − sin θ dy = 0
provides a model for a neighbourhood of S 1 .
21

Proof of Theorem 2.27. Choose contact forms αi for ξi, i = 0, 1, scaled in such a
way that Φ is actually an isomorphism of symplectic vector bundles with respect
to the symplectic bundle structures on CSN(Mi, Li) given by dαi. Here we think
of CSN(Mi, Li) as a sub-bundle of TMi|Li (rather than as a quotient bundle).
We identify (TMi|Li )/(ξi|Li ) with the trivial line bundle spanned by the Reeb
vector field Rαi . In total, this identifies
as a sub-bundle of TMi|Li .
NLi = 〈Rαi 〉 ⊕ Ji(TLi) ⊕ CSN(Mi, Li)
Let ΦR: 〈Rα0 〉 → 〈Rα1 〉 be the obvious bundle isomorphism defined by requiring
that Rα0 (p) map to Rα1 (φ(p)).
Let Ψi: Ji(TLi) → T ∗Li be the isomorphism defined by taking the interior
product with dαi. Notice that
Tφ ⊕ (φ ∗ ) −1 : (TL0 ⊕ T ∗ L0, ΩL0 ) → (TL1 ⊕ T ∗ L1, ΩL1 )
is an isomorphism of symplectic vector bundles. With Lemma 2.26 it follows that
Tφ ⊕ Ψ −1
1 ◦ (φ∗ ) −1 ◦ Ψ0: (TL0 ⊕ J0(TL0), dα0) → (TL1 ⊕ J1(TL1), dα1)
is an isomorphism of symplectic vector bundles.
Now let
�Φ: NL0 −→ NL1
be the bundle isomorphism (covering φ) defined by
�Φ = ΦR ⊕ Ψ −1
1 ◦ (φ∗ ) −1 ◦ Ψ0 ⊕ Φ.
Let τi: NLi → Mi be tubular maps, that is, the τ (I suppress the index i for
better readability) are embeddings such that τ|L – where L is identified with the
zero section of NL – is the inclusion L ⊂ M, and Tτ induces the identity on NL
along L (with respect to the splittings T(NL)|L = TL ⊕ NL = TM|L).
Then τ1 ◦ � Φ ◦ τ −1
0 : N(L0) → N(L1) is a diffeomorphism of suitable neighbourhoods
N(Li) of Li that induces the bundle map
Tφ ⊕ � Φ: TM0|L0 −→ TM1|L1 .
By construction, this bundle map pulls α1 back to α0 and dα1 to dα0. Hence, α0
and (τ1 ◦ � Φ ◦ τ −1
0 )∗α1 are contact forms on N(L0) that coincide on TM0|L0 , and
so do their differentials.
22

Now consider the family of 1–forms
βt = (1 − t)α0 + t(τ1 ◦ � Φ ◦ τ −1
0 )∗ α1, t ∈ [0, 1].
On TM0|L0 we have βt ≡ α0 and dβt ≡ dα0. Since the contact condition α ∧
(dα) n �= 0 is an open condition, we may assume – shrinking N(L0) if necessary
– that βt is a contact form on N(L0) for all t ∈ [0, 1]. By the Gray stability
theorem (Thm. 2.20) and Remark 2.21 (3) following its proof, we find an isotopy
ψt of N(L0), fixing L0, such that ψ ∗ t βt = λtα0 for some smooth family of smooth
functions λt: N(L0) → R + .
(Since N(L0) is not a closed manifold, ψt is a priori only a local flow. But
on L0 it is stationary and hence defined for all t. As in the proof of the Darboux
theorem (Thm. 2.24) we conclude that ψt is defined for all t ∈ [0, 1] in a sufficiently
small neighbourhood of L0, so shrinking N(L0) once again, if necessary, will
ensure that ψt is a global flow on N(L0).)
We conclude that ψ = τ1 ◦ � Φ ◦ τ −1
0 ◦ ψ1 is the desired contactomorphism.
Remark 2.30. With a little more care one can actually achieve Tψ1 = id on
TM0|L0 , which implies in particular that Tψ| CSN(M0,L0) = Φ, cf. [105]. (Remem-
ber that there is a certain freedom in constructing an isotopy via the Moser trick
if the condition Xt ∈ ξt is dropped.) The key point is the generalised Poincaré
lemma, cf. [80, p. 361], which allows us to write a closed differential form γ given
in a neighbourhood of the zero section of a bundle and vanishing along that zero
section as an exact form γ = dη with η and its partial derivatives with respect
to all coordinates (in any chart) vanishing along the zero section. This lemma is
applied first to γ = d(β1 − β0), in order to find (with the symplectic Moser trick)
a diffeomorphism σ of a neighbourhood of L0 ⊂ M0 with Tσ = id on TM0|L0
and such that dβ0 = d(σ ∗ β1). It is then applied once again to γ = β0 − σ ∗ β1.
(The proof of the symplectic neighbourhood theorem in [92] appears to be
incomplete in this respect.)
Example 2.31. Let M0 = M1 = R 3 with contact forms α0 = dz + x dy and
α1 = dz + (x + y)dy and L0 = L1 = 0 the origin in R 3 . Thus
We take Φ = idCSN.
CSN(M0, L0) = CSN(M1, L1) = span{∂x, ∂y} ⊂ T0R 3 .
23

Set αt = dz+(x+ty)dy. The Moser trick with Xt ∈ ker αt yields Xt = −y∂x,
and hence ψt(x, y, z) = (x − ty, y, z). Then
Tψ1 =
⎛
⎜
⎝
which does not restrict to Φ on CSN.
1 −1 0
0 1 0
0 0 1
⎞
⎟
⎠,
However, a different solution for ψ ∗ t αt = α0 is ψt(x, y, z) = (x, y, z − ty 2 /2),
found by integrating Xt = −y 2 ∂z/2 (a multiple of the Reeb vector field of αt).
Here we get
Tψ1 =
⎛
⎜
⎝
1 0 0
0 1 0
0 −y 1
⎞
⎟
⎠ ,
hence Tψ1| T0R 3 = id, so in particular Tψ1|CSN = Φ.
2.4.3 Contact submanifolds
Let (M ′ , ξ ′ = kerα ′ ) ⊂ (M, ξ = kerα) be a contact submanifold, that is, TM ′ ∩
ξ|M ′ = ξ′ . As before we write (ξ ′ ) ⊥ ⊂ ξ|M ′ for the symplectically orthogonal
complement of ξ ′ in ξ|M ′. Since M ′ is a contact submanifold (so ξ ′ is a symplectic
sub-bundle of (ξ|M ′, dα)), we have
TM ′ ⊕ (ξ ′ ) ⊥ = TM|M ′,
i.e. we can identify (ξ ′ ) ⊥ with the normal bundle NM ′ . Moreover, dα induces a
conformal symplectic structure on (ξ ′ ) ⊥ , so we call (ξ ′ ) ⊥ the conformal sym-
plectic normal bundle of M ′ in M and write
CSN(M, M ′ ) = (ξ ′ ) ⊥ .
Theorem 2.32. Let (Mi, ξi), i = 0, 1, be contact manifolds with compact contact
submanifolds (M ′ i , ξ′ i ). Suppose there is an isomorphism of conformal symplectic
normal bundles Φ: CSN(M0, M ′ 0 ) → CSN(M1, M ′ 1 ) that covers a contactomorphism
φ: (M ′ 0 , ξ′ 0 ) → (M ′ 1 , ξ′ 1 ). Then φ extends to a contactomorphism ψ of
suitable neighbourhoods N(M ′ i ) of M ′ i such that Tψ| CSN(M0,M ′ 0 ) and Φ are bundle
homotopic (as symplectic bundle isomorphisms) up to a conformality.
24

Example 2.33. A particular instance of this theorem is the case of a transverse
knot in a contact manifold (M, ξ), i.e. an embedding S 1 ֒→ (M, ξ) transverse to ξ.
Since the symplectic group Sp(2n) of linear transformations of R 2n preserving the
standard symplectic structure ω0 = � n
i=1 dxi ∧dyi is connected, there is only one
conformal symplectic R 2n –bundle over S 1 up to conformal equivalence. A model
for the neighbourhood of a transverse knot is given by
� S 1 × R 2n , ξ = ker � dθ +
n�
(xi dyi − yi dxi) �� ,
where θ denotes the S 1 –coordinate; the theorem says that in suitable local coor-
dinates the neighbourhood of any transverse knot looks like this model.
Proof of Theorem 2.32. As in the proof of Theorem 2.27 it is sufficient to find
contact forms αi on Mi and a bundle map TM0| M ′ 0 → TM1| M ′ 1 , covering φ and
inducing Φ, that pulls back α1 to α0 and dα1 to dα0; the proof then concludes
as there with a stability argument.
i=1
For this we need to make a judicious choice of αi. The essential choice is made
separately on each Mi, so I suppress the subscript i for the time being. Choose a
contact form α ′ for ξ ′ on M ′ . Write R ′ for the Reeb vector field of α ′ . Given any
contact form α for ξ on M we may first scale it such that α(R ′ ) ≡ 1 along M ′ .
Then α|TM ′ = α′ , and hence dα|TM ′ = dα′ . We now want to scale α further
such that its Reeb vector field R coincides with R ′ along M ′ . To this end it is
sufficient to find a smooth function f : M → R + with f|M ′ ≡ 1 and iR ′d(fα) ≡ 0
on TM|M ′. This last equation becomes
0 = iR ′d(fα) = iR ′(df ∧ α + f dα) = −df + iR ′dα on TM|M ′.
Since iR ′dα|TM ′ = iR ′dα′ ≡ 0, such an f can be found.
The choices of α ′ 0 and α′ 1 cannot be made independently of each other; we may
first choose α ′ 1 , say, and then define α′ 0 = φ∗α ′ 1 . Then define α0, α1 as described
and scale Φ such that it is a symplectic bundle isomorphism of
Then
((ξ ′ 0) ⊥ , dα0) −→ ((ξ ′ 1) ⊥ , dα1).
Tφ ⊕ Φ: TM0| M ′ 0 −→ TM1| M ′ 1
is the desired bundle map that pulls back α1 to α0 and dα1 to dα0.
25

Remark 2.34. The condition that Ri ≡ R ′ i along M ′ is necessary for ensuring
that (Tφ ⊕ Φ)(R0) = R1, which guarantees (with the other stated conditions)
that (Tφ ⊕ Φ) ∗ (dα1) = dα0. The condition dαi| TM ′ i = dα ′ i
choice of Φ alone would only give (Tφ ⊕ Φ) ∗ (dα1|ξ1 ) = dα0|ξ0 .
2.4.4 Hypersurfaces
and the described
Let S be an oriented hypersurface in a contact manifold (M, ξ = kerα) of dimen-
sion 2n + 1. In a neighbourhood of S in M, which we can identify with S × R
(and S with S × {0}), the contact form α can be written as
α = βr + ur dr,
where βr, r ∈ R, is a smooth family of 1–forms on S and ur : S → R a smooth
family of functions. The contact condition α ∧ (dα) n �= 0 then becomes
0 �= α ∧ (dα) n = (βr + ur dr) ∧ (dβr − ˙ βr ∧ dr + dur ∧ dr) n
= (−nβr ∧ ˙ βr + nβr ∧ dur + ur dβr) ∧ (dβr) n−1 ∧ dr, (2.5)
where the dot denotes derivative with respect to r. The intersection TS ∩ (ξ|S)
determines a distribution (of non-constant rank) of subspaces of TS. If α is
written as above, this distribution is given by the kernel of β0, and hence, at
a given p ∈ S, defines either the full tangent space TpS (if β0,p = 0) or a 1–
codimensional subspace both of TpS and ξp (if β0,p �= 0). In the former case, the
symplectically orthogonal complement (TpS ∩ξp) ⊥ (with respect to the conformal
symplectic structure dα on ξp) is {0}; in the latter case, (TpS ∩ ξp) ⊥ is a 1–
dimensional subspace of ξp contained in TpS ∩ ξp.
From that it is intuitively clear what one should mean by a ‘singular 1–
dimensional foliation’, and we make the following somewhat provisional defini-
tion.
Definition 2.35. The characteristic foliation Sξ of a hypersurface S in (M, ξ)
is the singular 1–dimensional foliation of S defined by (TS ∩ ξ|S) ⊥ .
Example 2.36. If dimM = 3 and dimS = 2, then (TpS ∩ξp) ⊥ = TpS ∩ξp at the
points p ∈ S where TpS ∩ ξp is 1–dimensional. Figure 2 shows the characteristic
foliation of the unit 2–sphere in (R 3 , ξ2), where ξ2 denotes the standard contact
26

structure of Example 2.7: The only singular points are (0, 0, ±1); away from these
points the characteristic foliation is spanned by
(y − xz)∂x − (x + yz)∂y + (x 2 + y 2 )∂z.
Figure 2: The characteristic foliation on S 2 ⊂ (R 3 , ξ2).
The following lemma helps to clarify the notion of singular 1–dimensional
foliation.
Lemma 2.37. Let β0 be the 1–form induced on S by a contact form α defining ξ,
and let Ω be a volume form on S. Then Sξ is defined by the vector field X
satisfying
iXΩ = β0 ∧ (dβ0) n−1 .
Proof. First of all, we observe that β0 ∧ (dβ0) n−1 �= 0 outside the zeros of β0:
Arguing by contradiction, assume β0,p �= 0 and β0 ∧(dβ0) n−1 |p = 0 at some p ∈ S.
Then (dβ0) n |p �= 0 by (2.5). On the codimension 1 subspace kerβ0,p of TpS the
symplectic form dβ0,p has maximal rank n−1. It follows that β0 ∧(dβ0) n−1 |p �= 0
after all, a contradiction.
Next we want to show that X ∈ ker β0. We observe
0 = iX(iXΩ) = β0(X)(dβ0) n−1 − (n − 1)β0 ∧ iXdβ0 ∧ (dβ0) n−2 . (2.6)
Taking the exterior product of this equation with β0 we get
β0(X)β0 ∧ (dβ0) n−1 = 0.
By our previous consideration this implies β0(X) = 0.
It remains to show that for β0,p �= 0 we have
dβ0(X(p), v) = 0 for all v ∈ TpS ∩ ξp.
27

For n = 1 this is trivially satisfied, because in that case v is a multiple of X(p).
I suppress the point p in the following calculation, where we assume n ≥ 2.
From (2.6) and with β0(X) = 0 we have
Taking the interior product with v ∈ TS ∩ ξ yields
β0 ∧ iXdβ0 ∧ (dβ0) n−2 = 0. (2.7)
−dβ0(X, v)β0 ∧ (dβ0) n−2 + (n − 2)β0 ∧ iXdβ0 ∧ ivdβ0 ∧ (dβ0) n−3 = 0.
(Thanks to the coefficient n − 2 the term (dβ0) n−3 is not a problem for n = 2.)
Taking the exterior product of that last equation with dβ0, and using (2.7), we
find
and thus dβ0(X, v) = 0.
dβ0(X, v)β0 ∧ (dβ0) n−1 = 0,
Remark 2.38. (1) We can now give a more formal definition of ‘singular 1–
dimensional foliation’ as an equivalence class of vector fields [X], where X is
allowed to have zeros and [X] = [X ′ ] if there is a nowhere zero function on all
of S such that X ′ = fX. Notice that the non-integrability of contact structures
and the reasoning at the beginning of the proof of the lemma imply that the zero
set of X does not contain any open subsets of S.
(2) If the contact structure ξ is cooriented rather than just coorientable, so
that α is well-defined up to multiplication with a positive function, then this
lemma allows to give an orientation to the characteristic foliation: Changing α
to λα with λ: M → R + will change β0 ∧ (dβ0) n−1 by a factor λ n .
We now restrict attention to surfaces in contact 3–manifolds, where the notion
of characteristic foliation has proved to be particularly useful.
The following theorem is due to E. Giroux [52].
Theorem 2.39 (Giroux). Let Si be closed surfaces in contact 3–manifolds (Mi, ξi),
i = 0, 1 (with ξi coorientable), and φ: S0 → S1 a diffeomorphism with φ(S0,ξ0 ) =
as oriented characteristic foliations. Then there is a contactomorphism
S1,ξ1
ψ: N(S0) → N(S1) of suitable neighbourhoods N(Si) of Si with ψ(S0) = S1
and such that ψ|S0 is isotopic to φ via an isotopy preserving the characteristic
foliation.
28

Proof. By passing to a double cover, if necessary, we may assume that the Si
are orientable hypersurfaces. Let αi be contact forms defining ξi. Extend φ to a
diffeomorphism (still denoted φ) of neighbourhoods of Si and consider the contact
forms α0 and φ ∗ α1 on a neighbourhood of S0, which we may identify with S0 ×R.
By rescaling α1 we may assume that α0 and φ ∗ α1 induce the same form β0
on S0 ≡ S0 × {0}, and hence also the same form dβ0.
Observe that the expression on the right-hand side of equation (2.5) is linear in
˙βr and ur. This implies that convex linear combinations of solutions of (2.5) (for
n = 1) with the same β0 (and dβ0) will again be solutions of (2.5) for sufficiently
small r. This reasoning applies to
αt := (1 − t)α0 + tφ ∗ α1, t ∈ [0, 1].
(I hope the reader will forgive the slight abuse of notation, with α1 denoting both
a form on M1 and its pull-back φ ∗ α1 to M0.) As is to be expected, we now use
the Moser trick to find an isotopy ψt with ψ ∗ t αt = λtα0, just as in the proof of
Gray stability (Theorem 2.20). In particular, we require as there that the vector
field Xt that we want to integrate to the flow ψt lie in the kernel of αt.
On TS0 we have ˙αt ≡ 0 (thanks to the assumption that α0 and φ ∗ α1 induce
the same form β0 on S0). In particular, if v is a vector in S0,ξ0 , then by equation
(2.1) we have dαt(Xt, v) = 0, which implies that Xt is a multiple of v, hence
tangent to S0,ξ0 . This shows that the flow of Xt preserves S0 and its characteristic
foliation. More formally, we have
LXtαt = d(αt(Xt)) + iXtdαt = iXtdαt,
so with v as above we have LXtαt(v) = 0, which shows that LXtαt|TS0 is a
multiple of α0|TS0 = β0. This implies that the (local) flow of Xt changes β0 by a
conformal factor.
Since S0 is closed, the local flow of Xt restricted to S0 integrates up to t = 1,
and so the same is true 3 in a neighbourhood of S0. Then ψ = φ ◦ ψ1 will be the
desired diffeomorphism N(S0) → N(S1).
As observed previously in the proof of Darboux’s theorem for contact forms,
the Moser trick allows more flexibility if we drop the condition αt(Xt) = 0.
We are now going to exploit this extra freedom to strengthen Giroux’s theorem
3 Cf. the proof (and the footnote therein) of Darboux’s theorem (Thm. 2.24).
29

slightly. This will be important later on when we want to extend isotopies of
hypersurfaces.
Theorem 2.40. Under the assumptions of the preceding theorem we can find
ψ: N(S0) → N(S1) satisfying the stronger condition that ψ|S0 = φ.
Proof. We want to show that the isotopy ψt of the preceding proof may be as-
sumed to fix S0 pointwise. As there, we may assume ˙αt|TS0
≡ 0.
If the condition that Xt be tangent to kerαt is dropped, the condition Xt
needs to satisfy so that its flow will pull back αt to λtα0 is
˙αt + d(αt(Xt)) + iXtdαt = µtαt, (2.8)
where µt and λt are related by µt = d
dt (log λt) ◦ ψ −1
t , cf. the proof of the Gray
stability theorem (Theorem 2.20).
Write Xt = HtRt+Yt with Rt the Reeb vector field of αt and Yt ∈ ξt = kerαt.
Then condition (2.8) translates into
˙αt + dHt + iYtdαt = µtαt. (2.9)
For given Ht one determines µt from this equation by inserting the Reeb vector
field Rt; the equation then admits a unique solution Yt ∈ kerαt because of the
non-degeneracy of dαt|ξt .
Our aim now is to ensure that Ht ≡ 0 on S0 and Yt ≡ 0 along S0. The latter
we achieve by imposing the condition
˙αt + dHt = 0 along S0
(2.10)
(which entails with (2.9) that µt|S0 ≡ 0). The conditions Ht ≡ 0 on S0 and (2.10)
can be simultaneously satisfied thanks to ˙αt|TS0 ≡ 0.
We can therefore find a smooth family of smooth functions Ht satisfying these
conditions, and then define Yt by (2.9). The flow of the vector field Xt = HtRt+Yt
then defines an isotopy ψt that fixes S0 pointwise (and thus is defined for all
t ∈ [0, 1] in a neighbourhood of S0). Then ψ = φ ◦ ψ1 will be the diffeomorphism
we wanted to construct.
2.4.5 Applications
Perhaps the most important consequence of the neighbourhood theorems proved
above is that they allow us to perform differential topological constructions such
30

as surgery or similar cutting and pasting operations in the presence of a contact
structure, that is, these constructions can be carried out on a contact manifold
in such a way that the resulting manifold again carries a contact structure.
One such construction that I shall explain in detail in Section 3 is the surgery
of contact 3–manifolds along transverse knots, which enables us to construct a
contact structure on every closed, orientable 3–manifold.
The concept of characteristic foliation on surfaces in contact 3–manifolds
has proved seminal for the classification of contact structures on 3–manifolds,
although it has recently been superseded by the notion of dividing curves. It is
clear that Theorem 2.39 can be used to cut and paste contact manifolds along
hypersurfaces with the same characteristic foliation. What actually makes this
useful in dimension 3 is that there are ways to manipulate the characteristic
foliation of a surface by isotoping that surface inside the contact 3–manifold.
The most important result in that direction is the Elimination Lemma proved
by Giroux [52]; an improved version is due to D. Fuchs, see [26]. This lemma
says that under suitable assumptions it is possible to cancel singular points of the
characteristic foliation in pairs by a C 0 –small isotopy of the surface (specifically:
an elliptic and a hyperbolic point of the same sign – the sign being determined
by the matching or non-matching of the orientation of the surface S and the
contact structure ξ at the singular point of Sξ). For instance, Eliashberg [24] has
shown that if a contact 3–manifold (M, ξ) contains an embedded disc D ′ such
that D ′ ξ
has a limit cycle, then one can actually find a so-called overtwisted disc:
an embedded disc D with boundary ∂D tangent to ξ (but D transverse to ξ
along ∂D, i.e. no singular points of Dξ on ∂D) and with Dξ having exactly one
singular point (of elliptic type); see Section 3.6.
Moreover, in the generic situation it is possible, given surfaces S ⊂ (M, ξ)
and S ′ ⊂ (M ′ , ξ ′ ) with Sξ homeomorphic to S ′ ξ ′, to perturb one of the surfaces so
as to get diffeomorphic characteristic foliations.
Chapter 8 of [1] contains a section on surfaces in contact 3–manifolds, and
in particular a proof of the Elimination Lemma. Further introductory reading
on the matter can be found in the lectures of J. Etnyre [35]; of the sources cited
above I recommend [26] as a starting point.
In [52] Giroux initiated the study of convex surfaces in contact 3–manifolds.
These are surfaces S with an infinitesimal automorphism X of the contact struc-
ture ξ with X transverse to S. For such surfaces, it turns out, much less infor-
31

mation than the characteristic foliation Sξ is needed to determine ξ in a neigh-
bourhood of S, viz., only the so-called dividing set of Sξ. This notion lies at the
centre of most of the recent advances in the classification of contact structures
on 3–manifolds [55], [71], [72]. A brief introduction to convex surface theory can
be found in [35].
2.5 Isotopy extension theorems
In this section we show that the isotopy extension theorem of differential topology
– an isotopy of a closed submanifold extends to an isotopy of the ambient manifold
– remains valid for the various distinguished submanifolds of contact manifolds.
The neighbourhood theorems proved above provide the key to the corresponding
isotopy extension theorems. For simplicity, I assume throughout that the ambient
contact manifold M is closed; all isotopy extension theorems remain valid if M has
non-empty boundary ∂M, provided the isotopy stays away from the boundary.
In that case, the isotopy of M found by extension keeps a neighbourhood of
∂M fixed. A further convention throughout is that our ambient isotopies ψt are
understood to start at ψ0 = idM.
2.5.1 Isotropic submanifolds
An embedding j : L → (M, ξ = kerα) is called isotropic if j(L) is an isotropic
submanifold of (M, ξ), i.e. everywhere tangent to the contact structure ξ. Equiv-
alently, one needs to require j ∗ α ≡ 0.
Theorem 2.41. Let jt: L → (M, ξ = kerα), t ∈ [0, 1], be an isotopy of isotropic
embeddings of a closed manifold L in a contact manifold (M, ξ). Then there is a
compactly supported contact isotopy ψt: M → M with ψt(j0(L)) = jt(L).
Proof. Define a time-dependent vector field Xt along jt(L) by
Xt ◦ jt = d
dt jt.
To simplify notation later on, we assume that L is a submanifold of M and j0 the
inclusion L ⊂ M. Our aim is to find a (smooth) family of compactly supported,
smooth functions � Ht: M → R whose Hamiltonian vector field � Xt equals Xt along
jt(L). Recall that � Xt is defined in terms of � Ht by
α( � Xt) = � Ht, i�Xt dα = d � Ht(Rα)α − d � Ht,
32

where, as usual, Rα denotes the Reeb vector field of α.
Hence, we need
α(Xt) = � Ht, iXtdα = d � Ht(Rα)α − d � Ht along jt(L). (2.11)
Write Xt = HtRα + Yt with Ht: jt(L) → R and Yt ∈ kerα. To satisfy (2.11) we
need
This implies
�Ht = Ht along jt(L). (2.12)
d � Ht(v) = dHt(v) for v ∈ T(jt(L)).
Since jt is an isotopy of isotropic embeddings, we have T(jt(L)) ⊂ ker α. So a
prerequisite for (2.11) is that
We have
dα(Xt, v) = −dHt(v) for v ∈ T(jt(L)). (2.13)
dα(Xt, v) + dHt(v) = dα(Xt, v) + d(α(Xt))(v)
so equation (2.13) is equivalent to
= iv(iXtdα + d(iXtα))
= iv(LXtα),
iv(LXtα) = 0 for v ∈ T(jt(L)).
But this is indeed tautologically satisfied: The fact that jt is an isotopy of isotropic
embeddings can be written as j ∗ t α ≡ 0; this in turn implies 0 = d
dt (j∗ t α) =
j ∗ t (LXtα), as desired.
This means that we can define � Ht by prescribing the value of � Ht along jt(L)
(with (2.12)) and the differential of � Ht along jt(L) (with (2.11)), where we are
free to impose d � Ht(Rα) = 0, for instance. The calculation we just performed
shows that these two requirements are consistent with each other. Any function
satisfying these requirements along jt(L) can be smoothed out to zero outside a
tubular neighbourhood of jt(L), and the Hamiltonian flow of this � Ht will be the
desired contact isotopy extending jt.
One small technical point is to ensure that the resulting family of functions
�Ht will be smooth in t. To achieve this, we proceed as follows. Set ˆ M = M ×[0, 1]
and
ˆL = �
q∈L,t∈[0,1]
33
(jt(q), t),

so that ˆ L is a submanifold of ˆ M. Let g be an auxiliary Riemannian metric on M
with respect to which Rα is orthogonal to kerα. Identify the normal bundle N ˆ L
of ˆ L in ˆ M with a sub-bundle of T ˆ M by requiring its fibre at a point (p, t) ∈ ˆ L
to be the g–orthogonal subspace of Tp(jt(L)) in TpM. Let τ : N ˆ L → ˆ M be a
tubular map.
Now define a smooth function ˆ H : N ˆ L → R as follows, where (p, t) always
denotes a point of ˆ L ⊂ N ˆ L.
• ˆ H(p, t) = α(Xt),
• d ˆ H (p,t)(Rα) = 0,
• d ˆ H (p,t)(v) = −dα(Xt, v) for v ∈ kerαp ⊂ TpM ⊂ T (p,t) ˆ M,
• ˆ H is linear on the fibres of N ˆ L → ˆ L.
Let χ: ˆ M → [0, 1] be a smooth function with χ ≡ 0 outside a small neighbour-
hood N0 ⊂ τ(N ˆ L) of ˆ L and χ ≡ 1 in a smaller neighbourhood N1 ⊂ N0 of ˆ L.
For (p, t) ∈ ˆ M, set
�Ht(p) =
�
χ(p, t) ˆ H(τ −1 (p, t)) for (p, t) ∈ τ(N ˆ L)
0 for (p, t) �∈ τ(N ˆ L).
This is smooth in p and t, and the Hamiltonian flow ψt of � Ht (defined globally
since � Ht is compactly supported) is the desired contact isotopy.
2.5.2 Contact submanifolds
An embedding j : (M ′ , ξ ′ ) → (M, ξ) is called a contact embedding if
is a contact submanifold of (M, ξ), i.e.
(j(M ′ ), Tj(ξ ′ ))
T(j(M)) ∩ ξ| j(M) = Tj(ξ ′ ).
If ξ = kerα, this can be reformulated as kerj ∗ α = ξ ′ .
Theorem 2.42. Let jt: (M ′ , ξ ′ ) → (M, ξ), t ∈ [0, 1], be an isotopy of con-
tact embeddings of the closed contact manifold (M ′ , ξ ′ ) in the contact manifold
(M, ξ). Then there is a compactly supported contact isotopy ψt: M → M with
ψt(j0(M ′ )) = jt(M ′ ).
34

Proof. In the proof of this theorem we follow a slightly different strategy from
the one in the isotropic case. Instead of directly finding an extension of the
Hamiltonian Ht: jt(M ′ ) → R, we first use the neighbourhood theorem for con-
tact submanifolds to extend jt to an isotopy of contact embeddings of tubular
neighbourhoods.
Again we assume that M ′ is a submanifold of M and j0 the inclusion M ′ ⊂ M.
As earlier, NM ′ denotes the normal bundle of M ′ in M. We also identify M ′
with the zero section of NM ′ , and we use the canonical identification
T(NM ′ )|M ′ = TM ′ ⊕ NM ′ .
By the usual isotopy extension theorem from differential topology we find an
isotopy
φt: NM ′ −→ M
with φt|M ′ = jt.
Choose contact forms α, α ′ defining ξ and ξ ′ , respectively. Define αt = φ∗ tα.
Then TM ′ ∩ kerαt = ξ ′ . Let R ′ denote the Reeb vector field of α ′ . Analogous
to the proof of Theorem 2.32, we first find a smooth family of smooth functions
gt: M ′ → R + such that gtαt|TM ′ = α′ , and then a family ft: NM ′ → R + with
ft|M ′ ≡ 1 and
dft = iR ′d(gtαt) on T(NM ′ )|M ′.
Then βt = ftgtαt is a family of contact forms on NM ′ representing the contact
structure ker(φ ∗ tα) and with the properties
βt|TM ′ = α′ ,
dβt|TM ′ = dα′ ,
ker(dβt) = 〈R ′ 〉 along M ′ .
The family (NM ′ , dβt) of symplectic vector bundles may be thought of as a
symplectic vector bundle over M ′ × [0, 1], which is necessarily isomorphic to a
bundle pulled back from M ′ × {0} (cf. [74, Cor. 3.4.4]). In other words, there is
a smooth family of symplectic bundle isomorphisms
Then
Φt: (NM ′ , dβ0) −→ (NM ′ , dβt).
idTM ′ ⊕ Φt: T(NM ′ )|M ′ −→ T(NM′ )|M ′
35

is a bundle map that pulls back βt to β0 and dβt to dβ0.
By the now familiar stability argument we find a smooth family of embeddings
ϕt: N(M ′ ) −→ NM ′
for some neighbourhood N(M ′ ) of the zero section M ′ in NM ′ with ϕ0 =
inclusion, ϕt|M ′ = idM ′ and ϕ∗ tβt = λtβ0, where λt: N(M ′ ) → R + . This means
that
φt ◦ ϕt: N(M ′ ) −→ M
is a smooth family of contact embeddings of (N(M ′ ), ker β0) in (M, ξ).
Define a time-dependent vector field Xt along φt ◦ ϕt(N(M ′ )) by
Xt ◦ φt ◦ ϕt = d
dt (φt ◦ ϕt).
This Xt is clearly an infinitesimal automorphism of ξ: By differentiating the
equation ϕ ∗ tφ ∗ tα = µtφ ∗ 0 α (where µt: N(M ′ ) → R + ) with respect to t we get
ϕ ∗ tφ ∗ t(LXtα) = ˙µtφ ∗ 0α = ˙µt
ϕ ∗ tφ ∗ tα,
so LXtα is a multiple of α (since φt ◦ ϕt is a diffeomorphism onto its image).
By the theory of contact Hamiltonians, Xt is the Hamiltonian vector field of
a Hamiltonian function ˆ Ht defined on φt ◦ ϕt(N(M ′ )). Cut off this function with
a bump function so as to obtain Ht: M → R with Ht ≡ ˆ Ht near φt ◦ ϕt(M ′ )
and Ht ≡ 0 outside a slightly larger neighbourhood of φt ◦ ϕt(M ′ ). Then the
Hamiltonian flow ψt of Ht satisfies our requirements.
2.5.3 Surfaces in 3–manifolds
Theorem 2.43. Let jt: S → (M, ξ = kerα), t ∈ [0, 1], be an isotopy of embed-
dings of a closed surface S in a 3–dimensional contact manifold (M, ξ). If all jt
induce the same characteristic foliation on S, then there is a compactly supported
isotopy ψt: M → M with ψt(j0(S)) = jt(S).
Proof. Extend jt to a smooth family of embeddings φt: S ×R → M, and identify
S with S × {0}. The assumptions say that all φ ∗ tα induce the same characteristic
foliation on S. By the proof of Theorem 2.40 and in analogy with the proof of
Theorem 2.42 we find a smooth family of embeddings
µt
ϕt: S × (−ε, ε) −→ S × R
36

for some ε > 0 with ϕ0 = inclusion, ϕt| S×{0} = idS and ϕ∗ tφ∗ tα = λtφ∗ 0α, where
λt: S × (−ε, ε) → R + . In other words, φt ◦ ϕt is a smooth family of contact
embeddings of (S × (−ε, ε), ker φ∗ 0α) in (M, ξ).
The proof now concludes exactly as the proof of Theorem 2.42.
2.6 Approximation theorems
A further manifestation of the (local) flexibility of contact structures is the fact
that a given submanifold can, under fairly weak (and usually obvious) topological
conditions, be approximated (typically C 0 –closely) by a contact submanifold or
an isotropic submanifold, respectively. The most general results in this direction
are best phrased in M. Gromov’s language of h-principles. For a recent text on
h-principles that puts particular emphasis on symplectic and contact geometry
see [30]; a brief and perhaps more gentle introduction to h-principles can be found
in [47].
In the present section I restrict attention to the 3–dimensional situation, where
the relevant approximation theorems can be proved by elementary geometric ad
hoc techniques.
Theorem 2.44. Let γ : S 1 → (M, ξ) be a knot, i.e. an embedding of S 1 , in
a contact 3–manifold. Then γ can be C 0 –approximated by a Legendrian knot
isotopic to γ. Alternatively, it can be C 0 –approximated by a positively as well as
a negatively transverse knot.
In order to prove this theorem, we first consider embeddings γ : (a, b) →
(R 3 , ξ) of an open interval in R 3 with its standard contact structure ξ = kerα,
where α = dz + x dy. Write γ(t) = (x(t), y(t), z(t)). Then
α(˙γ) = ˙z + x ˙y,
so the condition for a Legendrian curve reads ˙z + x ˙y ≡ 0; for a positively (resp.
negatively) transverse curve, ˙z + x ˙y > 0 (resp. < 0).
There are two ways to visualise such curves. The first is via its front pro-
jection
γF(t) = (y(t), z(t)),
the second via its Lagrangian projection
γL(t) = (x(t), y(t)).
37

2.6.1 Legendrian knots
If γ(t) = (x(t), y(t), z(t)) is a Legendrian curve in R 3 , then ˙y = 0 implies ˙z = 0,
so there the front projection has a singular point. Indeed, the curve t ↦→ (t, 0, 0)
is an example of a Legendrian curve whose front projection is a single point. We
call a Legendrian curve generic if ˙y = 0 only holds at isolated points (which we
call cusp points), and there ¨y �= 0.
Lemma 2.45. Let γ : (a, b) → (R 3 , ξ) be a Legendrian immersion. Then its front
projection γF(t) = (y(t), z(t)) does not have any vertical tangencies. Away from
the cusp points, γ is recovered from its front projection via
x(t) = − ˙z(t)
= −dz
˙y(t) dy ,
i.e. x(t) is the negative slope of the front projection. The curve γ is embedded if
and only if γF has only transverse self-intersections.
By a C ∞ –small perturbation of γ we can obtain a generic Legendrian curve ˜γ
isotopic to γ; by a C 2 –small perturbation we may achieve that the front projection
has only semi-cubical cusp singularities, i.e. around a cusp point at t = 0 the curve
˜γ looks like
with λ �= 0, see Figure 3.
˜γ(t) = (t + a, λt 2 + b, −λ(2t 3 /3 + at 2 ) + c)
Any regular curve in the (y, z)–plane with semi-cubical cusps and no vertical
tangencies can be lifted to a unique Legendrian curve in R 3 .
Figure 3: The cusp of a front projection.
Proof. The Legendrian condition is ˙z + x ˙y = 0. Hence ˙y = 0 forces ˙z = 0, so γF
cannot have any vertical tangencies.
Away from the cusp points, the Legendrian condition tells us how to recover
x as the negative slope of the front projection. (By continuity, the equation
38

x = dz
dy
also makes sense at generic cusp points.) In particular, a self-intersecting
front projection lifts to a non-intersecting curve if and only if the slopes at the
intersection point are different, i.e. if and only if the intersection is transverse.
That γ can be approximated in the C ∞ –topology by a generic immersion
˜γ follows from the usual transversality theorem (in its most simple form, viz.,
applied to the function y(t); the function x(t) may be left unchanged, and the
new z(t) is then found by integrating the new −x ˙y).
At a cusp point of ˜γ we have ˙y = ˙z = 0. Since ˜γ is an immersion, this forces
˙x �= 0, so ˜γ can be parametrised around a cusp point by the x–coordinate, i.e. we
may choose the curve parameter t such that the cusp lies at t = 0 and x(t) = t+a.
Since ¨y(0) �= 0 by the genericity condition, we can write y(t) = t 2 g(t) + y(0)
with a smooth function g(t) satisfying g(0) �= 0 (This is proved like the ‘Morse
lemma’ in Appendix 2 of [77]). A C 0 –approximation of g(t) by a function h(t)
with h(t) ≡ g(0) for t near zero and h(t) ≡ g(t) for |t| greater than some small
ε > 0 yields a C 2 –approximation of y(t) with the desired form around the cusp
point.
Example 2.46. Figure 4 shows the front projection of a Legendrian trefoil knot.
Figure 4: Front projection of a Legendrian trefoil knot.
Proof of Theorem 2.44 - Legendrian case. First of all, we consider a curve γ in
standard R 3 . In order to find a C 0 –close approximation of γ, we simply need
to choose a C 0 –close approximation of its front projection γF by a regular curve
without vertical tangencies and with isolated cusps (we call such a curve a front)
39

in such a way, that the slope of the front at time t is close to −x(t) (see Figure 5).
Then the Legendrian lift of this front is the desired C 0 –approximation of γ.
z
Figure 5: Legendrian C 0 –approximation via front projection.
If γ is defined on an interval (a, b) and is already Legendrian near its endpoints,
then the approximation of γF may be assumed to coincide with γF near the
endpoints, so that the Legendrian lift coincides with γ near the endpoints.
Hence, given a knot in an arbitrary contact 3–manifold, we can cut it (by the
Lebesgue lemma) into little pieces that lie in Darboux charts. There we can use
the preceding recipe to find a Legendrian approximation. Since, as just observed,
one can find such approximations on intervals with given boundary condition,
this procedure yields a Legendrian approximation of the full knot.
Locally (i.e. in R 3 ) the described procedure does not introduce any self-
intersections in the approximating curve, provided we approximate γF by a front
with only transverse self-intersections. Since the original knot was embedded, the
same will then be true for its Legendrian C 0 –approximation.
The same result may be derived using the Lagrangian projection:
Lemma 2.47. Let γ : (a, b) → (R 3 , ξ) be a Legendrian immersion. Then its
Lagrangian projection γL(t) = (x(t), y(t)) is also an immersed curve. The curve
γ is recovered from γL via
� t1
z(t1) = z(t0) −
40
t0
x dy.
y

A Legendrian immersion γ : S 1 → (R 3 , ξ) has a Lagrangian projection that en-
closes zero area. Moreover, γ is embedded if and only if every loop in γL (except,
in the closed case, the full loop γL) encloses a non-zero oriented area.
Any immersed curve in the (x, y)–plane is the Lagrangian projection of a
Legendrian curve in R 3 , unique up to translation in the z–direction.
Proof. The Legendrian condition ˙z + x ˙y implies that if ˙y = 0 then ˙z = 0, and
hence, since γ is an immersion, ˙x �= 0. So γL is an immersion.
The formula for z follows by integrating the Legendrian condition. For a
closed curve γL in the (x, y)–plane, the integral �
x dy computes the oriented
area enclosed by γL. From that all the other statements follow.
Example 2.48. Figure 6 shows the Lagrangian projection of a Legendrian un-
knot.
A −2A
Figure 6: Lagrangian projection of a Legendrian unknot.
Alternative proof of Theorem 2.44 – Legendrian case. Again we consider a curve
γ in standard R 3 defined on an interval. The generalisation to arbitrary contact
manifolds and closed curves is achieved as in the proof using front projections.
In order to find a C 0 –approximation of γ by a Legendrian curve, one only has
to approximate its Lagrangian projection γL by an immersed curve whose ‘area
integral’
z(t0) −
lies as close to the original z(t) as one wishes. This can be achieved by using small
loops oriented positively or negatively (see Figure 7). If γL has self-intersections,
this approximating curve can be chosen in such a way that along loops properly
contained in that curve the area integral is non-zero, so that again we do not
� t
t0
x dy
introduce any self-intersections in the Legendrian approximation of γ.
41
γL
A

y
Figure 7: Legendrian C 0 –approximation via Lagrangian projection.
2.6.2 Transverse knots
The quickest proof of the transverse case of Theorem 2.44 is via the Legendrian
case. However, it is perfectly feasible to give a direct proof along the lines of the
preceding discussion, i.e. using the front or the Lagrangian projection. Since this
picture is useful elsewhere, I include a brief discussion, restricting attention to
the front projection.
Thus, let γ(t) = (x(t), y(t), z(t)) be a curve in R 3 . The condition for γ to be
positively transverse to the standard contact structure ξ = ker(dz + x dy) is that
˙z + x ˙y > 0. Hence,
⎧
⎪⎨
⎪⎩
if ˙y = 0, then ˙z > 0,
if ˙y > 0, then x > − ˙z/ ˙y,
if ˙y < 0, then x < − ˙z/ ˙y.
The first statement says that there are no vertical tangencies oriented down-
wards in the front projection. The second statement says in particular that for
˙y > 0 and ˙z < 0 we have x > 0; the third, that for ˙y < 0 and ˙z < 0 we have
x < 0. This implies that the situations shown in Figure 8 are not possible in the
front projection of a positively transverse curve. I leave it to the reader to check
that all other oriented crossings are possible in the front projection of a positively
transverse curve, and that any curve in the (y, z)–plane without the forbidden
crossing or downward vertical tangencies admits a lift to a positively transverse
curve.
42
x

Figure 8: Impossible front projections of positively transverse curve.
Example 2.49. Figure 9 shows the front projection of a positively transverse
trefoil knot.
Figure 9: Front projection of a positively transverse trefoil knot.
Proof of Theorem 2.44 – transverse case. By the Legendrian case of this theo-
rem, the given knot γ can be C 0 –approximated by a Legendrian knot γ1. By
Example 2.29, a neighbourhood of γ1 in (M, ξ) looks like a solid torus S 1 × D 2
with contact structure cosθ dx − sinθ dy = 0, where γ1 ≡ S 1 × {0}. Then the
curve
γ2(t) = (θ = t, x = δ sint, y = δ cos t), t ∈ [0, 2π],
is a positively (resp. negatively) transverse knot if δ > 0 (resp. < 0). By choosing
|δ| small we obtain as good a C 0 –approximation of γ1 and hence of γ as we
wish.
3 Contact structures on 3–manifolds
Here is the main theorem proved in this section:
43

Theorem 3.1 (Lutz-Martinet). Every closed, orientable 3–manifold admits a
contact structure in each homotopy class of tangent 2–plane fields.
In Section 3.2 I present what is essentially J. Martinet’s [90] proof of the
existence of a contact structure on every 3–manifold. This construction is based
on a surgery description of 3–manifolds due to R. Lickorish and A. Wallace. For
the key step, showing how to extend over a solid torus certain contact forms
defined near the boundary of that torus (which then makes it possible to perform
Dehn surgeries), we use an approach due to W. Thurston and H. Winkelnkemper;
this allows to simplify Martinet’s proof slightly.
In Section 3.3 we show that every orientable 3–manifold is parallelisable and
then build on this to classify (co-)oriented tangent 2–plane fields up to homotopy.
In Section 3.4 we study the so-called Lutz twist, a topologically trivial Dehn
surgery on a contact manifold (M, ξ) which yields a contact structure ξ ′ on M
that is not homotopic (as 2–plane field) to ξ. We then complete the proof of the
main theorem stated above. These results are contained in R. Lutz’s thesis [84]
(which, I have to admit, I’ve never seen). Of Lutz’s published work, [83] only deals
with the 3–sphere (and is only an announcement); [85] also deals with a more
restricted problem. I learned the key steps of the construction from an exposition
given in V. Ginzburg’s thesis [50]. I have added proofs of many topological details
that do not seem to have appeared in a readily accessible source before.
In Section 3.5 I indicate two further proofs for the existence of contact struc-
tures on every 3–manifold (and provide references to others). The one by Thur-
ston and Winkelnkemper uses a description of 3–manifolds as open books due to
J. Alexander; the crucial idea in their proof is the one we also use to simplify
Martinet’s argument. Indeed, my discussion of the Lutz twist in the present
section owes more to the paper by Thurston-Winkelnkemper than to any other
reference. The second proof, by J. Gonzalo, is based on a branched cover descrip-
tion of 3–manifolds found by H. Hilden, J. Montesinos and T. Thickstun. This
branched cover description also yields a very simple geometric proof that every
orientable 3–manifold is parallelisable.
In Section 3.6 we discuss the fundamental dichotomy between tight and over-
twisted contact structures, introduced by Eliashberg, as well as the relation of
these types of contact structures with the concept of symplectic fillability. The
chapter concludes in Section 3.7 with a survey of classification results for contact
structures on 3–manifolds.
44

But first we discuss, in Section 3.1, an invariant of transverse knots in R 3
with its standard contact structure. This invariant will be an ingredient in the
proof of the Lutz-Martinet theorem, but is also of independent interest.
I do not feel embarrassed to use quite a bit of machinery from algebraic and
differential topology in this chapter. However, I believe that nothing that cannot
be found in such standard texts as [14], [77] and [95] is used without proof or an
explicit reference.
Throughout this third section, M denotes a closed, orientable 3-manifold.
3.1 An invariant of transverse knots
Although the invariant in question can be defined for transverse knots in arbitrary
contact manifolds (provided the knot is homologically trivial), for the sake of
clarity I restrict attention to transverse knots in R 3 with its standard contact
structure ξ0 = ker(dz + x dy). This will be sufficient for the proof of the Lutz-
Martinet theorem. In Section 3.7 I say a few words about the general situation
and related invariants for Legendrian knots.
Thus, let γ be a transverse knot in (R 3 , ξ0). Push γ a little in the direction
of ∂x – notice that this is a nowhere zero vector field contained in ξ0, and in
particular transverse to γ – to obtain a knot γ ′ . An orientation of γ induces an
orientation of γ ′ . The orientation of R 3 is given by dx ∧ dy ∧ dz.
Definition 3.2. The self-linking number l(γ) of the transverse knot γ is the
linking number of γ and γ ′ .
Notice that this definition is independent of the choice of orientation of γ (but
it changes sign if the orientation of R 3 is reversed). Furthermore, in place of ∂x
we could have chosen any nowhere zero vector field X in ξ0 to define l(γ): The
difference between the the self-linking number defined via ∂x and that defined
via X is the degree of the map γ → S 1 given by associating to a point on γ
the angle between ∂x and X. This degree is computed with the induced map
Z ∼ = H1(γ) → H1(S 1 ) ∼ = Z. But the map γ → S 1 factors through R 3 , so the
induced homomorphism on homology is the zero homomorphism.
Observe that l(γ) is an invariant under isotopies of γ within the class of
transverse knots.
We now want to compute l(γ) from the front projection of γ. Recall that the
writhe of an oriented knot diagram is the signed number of self-crossings of the
45

diagram, where the sign of the crossing is given in Figure 10.
−1
+1
Figure 10: Signs of crossings in a knot diagram.
Lemma 3.3. The self-linking number l(γ) of a transverse knot is equal to the
writhe w(γ) of its front projection.
Proof. Let γ ′ be the push-off of γ as described. Observe that each crossing of the
front projection of γ contributes a crossing of γ ′ underneath γ of the corresponding
sign. Since the linking number of γ and γ ′ is equal to the signed number of times
that γ ′ crosses underneath γ (cf. [98, p. 37]), we find that this linking number is
equal to the signed number of self-crossings of γ, that is, l(γ) = w(γ).
Proposition 3.4. Every self-linking number is realised by a transverse link in
standard R 3 .
Proof. Figure 11 shows front projections of positively transverse knots (cf. Sec-
tion 2.6.2) with self-linking number ±3. From that the construction principle for
realising any odd integer should be clear. With a two component link any even
integer can be realised.
Remark 3.5. It is no accident that I do not give an example of a transverse
knot with even self-linking number. By a theorem of Eliashberg [26, Prop. 2.3.1]
that relates l(γ) to the Euler characteristic of a Seifert surface S for γ and the
signed number of singular points of the characteristic foliation Sξ, the self-linking
number l(γ) of a knot can only take odd values.
3.2 Martinet’s construction
According to Lickorish [81] and Wallace [103] M can be obtained from S 3 by
Dehn surgery along a link of 1–spheres. This means that we have to remove
46

−3
+3
Figure 11: Transverse knots with self-linking number ±3.
a disjoint set of embedded solid tori S 1 × D 2 from S 3 and glue back solid tori
with suitable identification by a diffeomorphism along the boundaries S 1 × S 1 .
The effect of such a surgery (up to diffeomorphism of the resulting manifold) is
completely determined by the induced map in homology
H1(S 1 × ∂D 2 ) −→ H1(S 1 × ∂D 2 )
Z ⊕ Z −→ Z ⊕ Z,
� �
which is given by a unimodular matrix
n q
m p
∈ GL(2, Z). Hence, denoting
coordinates in S 1 × S 1 by (θ, ϕ), we may always assume the identification maps
to be of the form �
θ
ϕ
�
↦−→
�
n q
m p
The curves µ and λ on ∂(S 1 × D 2 ) given respectively by θ = 0 and ϕ = 0 are
called meridian and longitude. We keep the same notation µ, λ for the homology
classes these curves represent. It turns out that the diffeomorphism type of the
surgered manifold is completely determined by the class pµ + qλ, which is the
class of the curve that becomes homotopically trivial in the surgered manifold
(cf. [98, p. 28]). In fact, the Dehn surgery is completely determined by the surgery
coefficient p/q, since the diffeomorphism of ∂(S 1 ×D 2 ) given by (λ, µ) ↦→ (λ, −µ)
��
extends to a diffeomorphism of the solid torus that we glue back.
Similarly, the diffeomorphism given by (λ, µ) ↦→ (λ + kµ, µ) extends. By such
a change of longitude in S 1 × D 2 ⊂ M, which amounts to choosing a different
47
θ
ϕ
�
.

trivialisation of the normal bundle (i.e. framing) of S1 �
�
×{0} ⊂ M, the gluing map
is changed to
n q
m − kn p − kq
. By a change of longitude in the solid torus
� �
that we glue back, the gluing map is changed to
n + kq q
m + kp p
. Thus, a Dehn
surgery is a so-called handle surgery (or ‘honest surgery’ or simply ‘surgery’) if
and only if the surgery coefficient � is �an
integer, � that � is, q = ±1. For in exactly
this case we may assume
n q
m p
=
0 1
1 0
, and the surgery is given by
cutting out S 1 × D 2 and gluing back S 1 × D 2 with the identity map
∂(D 2 × S 1 ) −→ ∂(S 1 × D 2 ).
The theorem of Lickorish and Wallace remains true if we only allow handle
surgeries. However, this assumption does not entail any great simplification of
the existence proof for contact structures, so we shall work with general Dehn
surgeries.
Our aim in this section is to use this topological description of 3–manifolds
for a proof of the following theorem, first proved by Martinet [90]. The proof
presented here is in spirit the one given by Martinet, but, as indicated in the
introduction to this third section, amalgamated with ideas of Thurston and
Winkelnkemper [101], whose proof of the same theorem we shall discuss later.
Theorem 3.6 (Martinet). Every closed, orientable 3–manifold M admits a con-
tact structure.
In view of the theorem of Lickorish and Wallace and the fact that S 3 admits
a contact structure, Martinet’s theorem is a direct consequence of the following
result.
Theorem 3.7. Let ξ0 be a contact structure on a 3–manifold M0. Let M be the
manifold obtained from M0 by a Dehn surgery along a knot K. Then M admits a
contact structure ξ which coincides with ξ0 outside the neighbourhood of K where
we perform surgery.
Proof. By Theorem 2.44 we may assume that K is positively transverse to ξ0.
Then, by the contact neighbourhood theorem (Example 2.33), we can find a
tubular neighbourhood of K diffeomorphic to S 1 × D 2 (δ0), where K is identified
with S 1 × {0} and D 2 (δ0) denotes a disc of radius δ0, such that the contact
48

structure ξ0 is given as the kernel of dθ+r 2 dϕ, with θ denoting the S 1 –coordinate
and (r, ϕ) polar coordinates on D 2 (δ0).
� Now perform � a Dehn surgery along K defined by the unimodular matrix
n q
m p
. This corresponds to cutting out S 1 ×D 2 (δ1) ⊂ S 1 ×D 2 (δ0) for some
δ1 < δ0 and gluing it back in the way described above.
Write (θ; r, ϕ) for the coordinates on the copy of S 1 × D 2 (δ1) that we want
to glue back. Then the contact form dθ + r 2 dϕ given on S 1 × D 2 (δ0) pulls back
(along S 1 × ∂D 2 (δ1)) to
d(nθ + qϕ) + r 2 d(mθ + pϕ).
This form is defined on all of S 1 × (D 2 (δ1) − {0}), and to complete the proof it
only remains to find a contact form on S 1 × D 2 (δ1) that coincides with this form
near S 1 × ∂D 2 (δ1). It is at this point that we use an argument inspired by the
Thurston-Winkelnkemper proof (but which goes back to Lutz).
� �
Lemma 3.8. Given a unimodular matrix
n q
m p
, there is a contact form on
S 1 × D 2 (δ) that coincides with (n + mr 2 )dθ + (q + pr 2 )dϕ near r = δ and with
±dθ + r 2 dϕ near r = 0.
Proof. We make the ansatz
α = h1(r)dθ + h2(r)dϕ
with smooth functions h1(r), h2(r). Then
and
dα = h ′ 1 dr ∧ dθ + h ′ 2 dr ∧ dϕ
�
�
�
α ∧ dα = �
�
h1 h2
h ′ 1 h′ 2
�
�
�
� dθ ∧ dr ∧ dϕ.
�
So to satisfy the contact condition α ∧ dα �= 0 all we have to do is to find a
parametrised curve
r ↦−→ (h1(r), h2(r)), 0 ≤ r ≤ δ,
in the plane satisfying the following conditions:
1. h1(r) = ±1 and h2(r) = r 2 near r = 0,
49

2. h1(r) = n + mr 2 and h2(r) = q + pr 2 near r = δ,
3. (h1(r), h2(r)) is never parallel to (h ′ 1 (r), h′ 2 (r)).
Since np−mq = ±1, the vector (m, p) is not a multiple of (n, q). Figure 12 shows
possible solution curves for the two cases np − mq = ±1.
−1
h2
(n,q)
(n + m,q + p)
h1
h2
Figure 12: Dehn surgery.
(n + m,q + p)
1
(n,q)
This completes the proof of the lemma and hence that of Theorem 3.7.
Remark 3.9. On S 3 we have the standard contact forms α± = x dy − y dx ±
(z dt−t dz) defining opposite orientations (cf. Remark 2.2). Performing the above
surgery construction either on (S 3 , kerα+) or on (S 3 , kerα−) we obtain both
positive and negative contact structures on any given M. The same is true for
the Lutz construction that we study in the next two sections. Hence: A closed
oriented 3–manifold admits both a positive and a negative contact structure in
each homotopy class of tangent 2–plane fields.
3.3 2–plane fields on 3–manifolds
First we need the following well-known fact.
Theorem 3.10. Every closed, orientable 3–manifold M is parallelisable.
Remark. The most geometric proof of this theorem can be given based on a
structure theorem of Hilden, Montesinos and Thickstun. This will be discussed
in Section 3.5.2. Another proof can be found in [76]. Here we present the classical
algebraic proof.
50
h1

Proof. The main point is to show the vanishing of the second Stiefel-Whitney
class w2(M) = w2(TM) ∈ H 2 (M; Z2). Recall the following facts, which can
be found in [14]; for the interpretation of Stiefel-Whitney classes as obstruction
classes see also [95].
There are Wu classes vi ∈ H i (M; Z2) defined by
〈Sq i (u), [M]〉 = 〈vi ∪ u, [M]〉
for all u ∈ H 3−i (M; Z2), where Sq denotes the Steenrod squaring operations.
Since Sq i (u) = 0 for i > 3 − i, the only (potentially) non-zero Wu classes are
v0 = 1 and v1. The Wu classes and the Stiefel-Whitney classes are related by
wq = �
j Sqq−j (vj). Hence v1 = Sq 0 (v1) = w1, which equals zero since M is
orientable. We conclude w2 = 0.
Let V2(R 3 ) = SO(3)/SO(1) = SO(3) be the Stiefel manifold of oriented,
orthonormal 2–frames in R 3 . This is connected, so there exists a section over
the 1–skeleton of M of the 2–frame bundle V2(TM) associated with TM (with a
choice of Riemannian metric on M understood 4 ). The obstruction to extending
this section over the 2–skeleton is equal to w2, which vanishes as we have just seen.
The obstruction to extending the section over all of M lies in H 3 (M; π2(V2(R 3 ))),
which is the zero group because of π2(SO(3)) = 0.
We conclude that TM has a trivial 2–dimensional sub-bundle ε 2 . The com-
plementary 1–dimensional bundle λ = TM/ε 2 is orientable and hence trivial
since 0 = w1(TM) = w1(ε 2 ) + w1(λ) = w1(λ). Thus TM = ε 2 ⊕ λ is a trivial
bundle.
Fix an arbitrary Riemannian metric on M and a trivialisation of the unit
tangent bundle STM ∼ = M × S 2 . This sets up a one-to-one correspondence
between the following sets, where all maps, homotopies etc. are understood to be
smooth.
• Homotopy classes of unit vector fields X on M,
• Homotopy classes of (co-)oriented 2–plane distributions ξ in TM,
• Homotopy classes of maps f : M → S 2 .
4 This is not necessary, of course. One may also work with arbitrary 2–frames without refer-
ence to a metric. This does not affect the homotopical data.
51

(I use the term ‘2–plane distribution’ synomymously with ‘2–dimensional sub-
bundle of the tangent bundle’.) Let ξ1, ξ2 be two arbitrary 2–plane distributions
(always understood to be cooriented). By elementary obstruction theory there is
an obstruction
d 2 (ξ1, ξ2) ∈ H 2 (M; π2(S 2 )) ∼ = H 2 (M; Z)
for ξ1 to be homotopic to ξ2 over the 2–skeleton of M and, if d 2 (ξ1, ξ2) = 0 and
after homotoping ξ1 to ξ2 over the 2–skeleton, an obstruction (which will depend,
in general, on that first homotopy)
d 3 (ξ1, ξ2) ∈ H 3 (M; π3(S 2 )) ∼ = H 3 (M; Z) ∼ = Z
for ξ1 to be homotopic to ξ2 over all of M. (The identification of H 3 (M; Z) with Z
is determined by the orientation of M.) However, rather than relying on general
obstruction theory, we shall interpret d 2 and d 3 geometrically, which will later
allow us to give a geometric proof that every homotopy class of 2–plane fields ξ
on M contains a contact structure.
The only fact that I want to quote here is that, by the Pontrjagin-Thom
construction, homotopy classes of maps f : M → S 2 are in one-to-one correspon-
dence with framed cobordism classes of framed (and oriented) links of 1–spheres
in M. The general theory can be found in [14] and [77]; a beautiful and elemen-
tary account is given in [94].
For given f, the correspondence is defined by choosing a regular value p ∈ S 2
for f and a positively oriented basis b of TpS 2 , and associating with it the oriented
framed link (f −1 (p), f ∗ b), where f ∗ b is the pull-back of b under the fibrewise
bijective map Tf : T(f −1 (p)) ⊥ → TpS 2 . The orientation of f −1 (p) is the one
which together with the frame f ∗ b gives the orientation of M.
For a given framed link L the corresponding f is defined by projecting a
(trivial) disc bundle neighbourhood L × D 2 of L in M onto the fibre D 2 ∼ =
S 2 −p ∗ , where 0 is identified with p and p ∗ denotes the antipode of p, and sending
M − (L × D 2 ) to p ∗ . Notice that the orientations of M and the components of L
determine that of the fibre D 2 , and hence determine the map f.
Before proceeding to define the obstruction classes d 2 and d 3 we make a
short digression and discuss some topological background material which is fairly
standard but not contained in our basic textbook references [14] and [77].
52

3.3.1 Hopf’s Umkehrhomomorphismus
If f : M m → N n is a continuous map between smooth manifolds, one can define
a homomorphism ϕ : Hn−p(N) → Hm−p(M) on homology classes represented by
submanifolds as follows. Given a homology class [L]N ∈ Hn−p(N) represented by
a codimension p submanifold L, replace f by a smooth approximation transverse
to L and define ϕ([L]N) = [f −1 (L)]M. This is essentially Hopf’s Umkehrhomo-
morphismus [73], except that he worked with combinatorial manifolds of equal
dimension and made no assumptions on the homology class. The following theo-
rem, which in spirit is contained in [41], shows that ϕ is independent of choices (of
submanifold L representing a class and smooth transverse approximation to f)
and actually a homomorphism of intersection rings. This statement is not as well-
known as it should be, and I only know of a proof in the literature for the special
case where L is a point [60]. In [14] this map is called transfer map (more general
transfer maps are discussed in [60]), but is only defined indirectly via Poincaré
duality (though implicitly the statement of the following theorem is contained
in [14], see for instance p. 377).
Theorem 3.11. Let f : M m → N n be a smooth map between closed, oriented
manifolds and L n−p ⊂ N n a closed, oriented submanifold of codimension p such
that f is transverse to L. Write u ∈ H p (N) for the Poincaré dual of [L]N, that
is, u∩[N] = [L]N. Then [f −1 (L)]M = f ∗ u∩[M]. In other words: If u is Poincaré
dual to [L]N, then f ∗ u ∈ H p (M) is Poincaré dual to [f −1 (L)]M.
Proof. Since f is transverse to L, the differential Tf induces a fibrewise isomor-
phism between the normal bundles of f −1 (L) and L, and we find (closed) tubular
neighbourhoods W → L and V = f −1 (W) → f −1 (L) (considered as disc bun-
dles) such that f : V → W is a fibrewise isomorphism. Write [V ]0 and [W]0 for
the orientation classes in Hm(V, V − f −1 (L)) and Hn(W, W − L), respectively.
We can identify these homology groups with Hm(V, ∂V ) and Hn(W, ∂W), respec-
tively. Let τW ∈ H p (W, ∂W) and τV ∈ H p (V, ∂V ) be the Thom classes of these
disc bundles defined by
τW ∩ [W]0 = [L]N,
τV ∩ [V ]0 = [f −1 (L)]M.
Notice that f ∗ τW = τV since f : W → V is fibrewise isomorphic and the Thom
class of an oriented disc bundle is the unique class whose restriction to each fibre
53

is a positive generator of H p (D p , ∂D p ). Writing i: M → (M, M − f −1 (L)) and
j : N → (N, N − L) for the inclusion maps we have
[f −1 (L)]M = τV ∩ [V ]0
= f ∗ τW ∩ [V ]0
= f ∗ τW ∩ i∗[M],
where we identify Hm(M, M −f −1 (L)) with Hm(V, V −f −1 (L)) under the excision
isomorphism. Then we have further
[f −1 (L)]M = i ∗ f ∗ τW ∩ [M]
= f ∗ j ∗ τW ∩ [M].
So it remains to identify j ∗ τW as the Poincaré dual u of [L]N. Indeed,
j ∗ τW ∩ [N] = τW ∩ j∗[N]
= τW ∩ [W]0
= [L]N,
where we have used the excision isomorphism between the groups Hn(W, W −L)
and Hn(N, N − L).
3.3.2 Representing homology classes by submanifolds
We now want to relate elements in H1(M; Z) to cobordism classes of links in M.
Theorem 3.12. Let M be a closed, oriented 3–manifold. Any c ∈ H1(M; Z)
is represented by an embedded, oriented link (of 1–spheres) Lc in M. Two links
L0, L1 represent the same class [L0] = [L1] if and only if they are cobordant in M,
that is, there is an embedded, oriented surface S in M × [0, 1] with
where ⊔ denotes disjoint union.
∂S = L1 ⊔ (−L0) ⊂ M × {1} ⊔ M × {0},
Proof. Given c ∈ H1(M; Z), set u = PD(c) ∈ H 2 (M; Z), where PD denotes
the Poincaré duality map from homology to cohomology. There is a well-known
isomorphism
H 2 (M; Z) ∼ = [M, K(Z, 2)] = [M, CP ∞ ],
54

where brackets denote homotopy classes of maps (cf. [14, VII.12]). So u cor-
responds to a homotopy class of maps [f]: M → CP ∞ such that f ∗ u0 = u,
where u0 is the positive generator of H 2 (CP ∞ ) (that is, the one that pulls back
to the Poincaré dual of [CP k−1 ] CP k under the natural inclusion CP k ⊂ CP ∞ ).
Since dimM = 3, any map f : M → CP ∞ is homotopic to a smooth map
f1: M → CP 1 . Let p be a regular value of f1. Then
PD(c) = u = f ∗ 1u0 = f ∗ 1PD[p] = PD[f −1
1 (p)]
by our discussion above, and hence c = [f −1
link.
1 (p)]. So Lc = f −1
1
(p) is the desired
It is important to note that in spite of what we have just said it is not true that
[M, CP ∞ ] = [M, CP 1 ], since a map F : M ×[0, 1] → CP ∞ with F(M × {0, 1}) ⊂
CP 1 is not, in general, homotopic rel (M × {0, 1}) to a map into CP 1 . However,
we do have [M, CP ∞ ] = [M, CP 2 ].
If two links L0, L1 are cobordant in M, then clearly
[L0] = [L1] ∈ H1(M × [0, 1]; Z) ∼ = H1(M; Z).
For the converse, suppose we are given two links L0, L1 ⊂ M with [L0] = [L1].
Choose arbitrary framings for these links and use this, as described at the be-
ginning of this section, to define smooth maps f0, f1: M → S 2 with common
regular value p ∈ S 2 such that f −1
i (p) = Li, i = 0, 1. Now identify S 2 with the
standardly embedded CP 1 ⊂ CP 2 . Let P ⊂ CP 2 be a second copy of CP 1 , em-
bedded in such a way that [P] CP 2 = [CP 1 ] CP 2 and P intersects CP 1 transversely
in p only. This is possible since CP 1 ⊂ CP 2 has self-intersection one. Then
the maps f0, f1, regarded as maps into CP 2 , are transverse to P and we have
f −1
i (P) = Li, i = 0, 1. Hence
f ∗ i u0 = f ∗ i (PD[P] CP 2) = PD[f −1
i (P)]M
= PD[Li]M
is the same for i = 0 or 1, and from the identification
[M, CP 2 ]
∼=
−→ H 2 (M, Z)
[f] ↦−→ f ∗ u0
we conclude that f0 and f1 are homotopic as maps into CP 2 .
55

Let F : M × [0, 1] → CP 2 be a homotopy between f0 and f1, which we may
assume to be constant near 0 and 1. This F can be smoothly approximated by a
map F ′ : M × [0, 1] → CP 2 which is transverse to P and coincides with F near
M ×0 and M ×1 (since there the transversality condition was already satisfied).
In particular, F ′ is still a homotopy between f0 and f1, and S = (F ′ ) −1 (P) is a
surface with the desired property ∂S = L1 ⊔ (−L0).
Notice that in the course of this proof we have observed that cobordism classes
of links in M (equivalently, classes in H1(M; Z)) correspond to homotopy classes
of maps M → CP 2 , whereas framed cobordism classes of framed links correspond
to homotopy classes of maps M → CP 1 .
By forming the connected sum of the components of a link representing a
certain class in H1(M; Z), one may actually always represent such a class by a
link with only one component, that is, a knot.
3.3.3 Framed cobordisms
We have seen that if L1, L2 ⊂ M are links with [L1] = [L2] ∈ H1(M; Z), then
L1 and L2 are cobordant in M. In general, however, a given framing on L1 and
L2 does not extend over the cobordism. The following observation will be useful
later on.
Write (S 1 , n) for a contractible loop in M with framing n ∈ Z (by which
we mean that S 1 and a second copy of S 1 obtained by pushing it away in the
direction of one of the vectors in the frame have linking number n). When writing
L = L ′ ⊔ (S 1 , n) it is understood that (S 1 , n) is not linked with any component
of L ′ .
Suppose we have two framed links L0, L1 ⊂ M with [L0] = [L1]. Let S ⊂
M × [0, 1] be an embedded surface with
∂S = L1 ⊔ (−L0) ⊂ M × {1} ⊔ M × {0}.
With D 2 a small disc embedded in S, the framing of L1 and L2 in M extends
to a framing of S − D 2 in M × [0, 1] (since S − D 2 deformation retracts to a 1–
dimensional complex containing L0 and L1, and over such a complex an orientable
2–plane bundle is trivial). Now we embed a cylinder S 1 ×[0, 1] in M ×[0, 1] such
that
S 1 × [0, 1] ∩ M × {0} = ∅,
56

and
S 1 × [0, 1] ∩ M × {1} = S 1 × {1},
S 1 × [0, 1] ∩ (S − D 2 ) = S 1 × {0} = ∂D 2 ,
see Figure 13. This shows that L0 is framed cobordant in M to L1 ⊔ (S 1 , n) for
suitable n ∈ Z.
L0 ⊂ M × {0}
D 2
S
S 1 × [0,1]
L1 ⊂ M × {1}
Figure 13: The framed cobordism between L0 and L1 ⊔ (S 1 , n).
3.3.4 Definition of the obstruction classes
We are now in a position to define the obstruction classes d 2 and d 3 . With a
choice of Riemannian metric on M and a trivialisation of STM understood, a
2–plane distribution ξ on M defines a map fξ : M → S 2 and hence an oriented
framed link Lξ as described above. Let [Lξ] ∈ H1(M; Z) be the homology class
represented by Lξ. This only depends on the homotopy class of ξ, since under
homotopies of ξ or choice of different regular values of fξ the cobordism class of
Lξ remains invariant. We define
d 2 (ξ1, ξ2) = PD[Lξ1 ] − PD[Lξ2 ].
With this definition d 2 is clearly additive, that is,
d 2 (ξ1, ξ2) + d 2 (ξ2, ξ3) = d 2 (ξ1, ξ3).
The following lemma shows that d 2 is indeed the desired obstruction class.
57

Lemma 3.13. The 2–plane distributions ξ1 and ξ2 are homotopic over the 2–
skeleton M (2) of M if and only if d 2 (ξ1, ξ2) = 0.
Proof. Suppose d2 (ξ1, ξ2) = 0, that is, [Lξ1 ] = [Lξ2 ]. By Theorem 3.12 we find a
surface S in M × [0, 1] with
∂S = Lξ2 ⊔ (−Lξ1 ) ⊂ M × {1} ⊔ M × {0}.
From the discussion on framed cobordism above we know that for suitable n ∈ Z
we find a framed surface S ′ in M × [0, 1] such that
as framed manifolds.
∂S ′ = � Lξ2 ⊔ (S1 , n) � ⊔ (−Lξ1 ) ⊂ M × {1} ⊔ M × {0}
Hence ξ1 is homotopic to a 2–plane distribution ξ ′ 1 such that L ξ ′ 1
and Lξ2 differ
only by one contractible framed loop (not linked with any other component).
Then the corresponding maps f ′ 1 , f2 differ only in a neighbourhood of this loop,
which is contained in a 3–ball, so f ′ 1 and f2 (and hence ξ ′ 1 and ξ2) agree over the
2–skeleton.
Conversely, if ξ1 and ξ2 are homotopic over M (2) , we may assume ξ1 = ξ2 on
M − D3 for some embedded 3–disc D3 ⊂ M without changing [Lξ1 ] and [Lξ2 ].
Now [Lξ1 ] = [Lξ2 ] follows from H1(D3 , S2 ) = 0.
Remark 3.14. By [99, § 37] the obstruction to homotopy between ξ and ξ0
(corresponding to the constant map fξ0 : M → S2 ) over the 2–skeleton of M is
given by f ∗ ξ u0, where u0 is the positive generator of H 2 (S 2 ; Z). So u0 = PD[p]
for any p ∈ S 2 , and taking p to be a regular value of fξ we have
f ∗ ξ u0 = f ∗ −1
ξ PD[p] = PD[fξ (p)]
= PD[Lξ] = d 2 (ξ, ξ0).
This gives an alternative way to see that our geometric definition of d 2 does
indeed coincide with the class defined by classical obstruction theory.
Now suppose d 2 (ξ1, ξ2) = 0. We may then assume that ξ1 = ξ2 on M−int(D 3 ),
and we define d 3 (ξ1, ξ2) to be the Hopf invariant H(f) of the map f : S 3 → S 2
defined as f1 ◦ π+ on the upper hemisphere and f2 ◦ π− on the lower hemisphere,
where π+, π− are the orthogonal projections of the upper resp. lower hemisphere
58

onto the equatorial disc, which we identify with D 3 ⊂ M. Here, given an orien-
tation of M, we orient S 3 in such a way that π+ is orientation-preserving and
π− orientation-reversing; the orientation of S 2 is inessential for the computation
of H(f). Recall that H(f) is defined as the linking number of the preimages of
two distinct regular values of a smooth map homotopic to f. Since the Hopf in-
variant classifies homotopy classes of maps S 3 → S 2 (it is in fact an isomorphism
π3(S 2 ) → Z), this is a suitable definition for the obstruction class d 3 . Moreover,
the homomorphism property of H(f) and the way addition in π3(S 2 ) is defined
entail the additivity of d 3 analogous to that of d 2 .
For M = S 3 there is another way to interpret d 3 . Oriented 2–plane distrib-
utions on M correspond to sections of the bundle associated to TM with fibre
SO(3)/U(1), hence to maps M → SO(3)/U(1) ∼ = S 2 since TM is trivial. Simi-
larly, almost complex structures on D 4 correspond to maps D 4 → SO(4)/U(2) ∼ =
SO(3)/U(1) (cf. [61] for this isomorphism). A cooriented 2–plane distribution on
M can be interpreted as a triple (X, ξ, J), where X is a vector field transverse
to ξ defining the coorientation, and J a complex structure on ξ defining the ori-
entation. Such a triple is called an almost contact structure. (This notion
generalises to higher (odd) dimensions, and by Remark 2.3 every cooriented con-
tact structure induces an almost contact structure, and in fact a unique one up
to homotopy as follows from the result cited in that remark.) Given an almost
contact structure (X, ξ, J) on S 3 , one defines an almost complex structure � J on
TD 4 |S 3 by � J|ξ = J and � JN = X, where N denotes the outward normal vector
field. So there is a canonical way to identify homotopy classes of almost con-
tact structures on S 3 with elements of π3(SO(3)/U(1)) ∼ = Z such that the value
zero corresponds to the almost contact structure that extends as almost complex
structure over D 4 .
3.4 Let’s Twist Again
Consider a 3–manifold M with cooriented contact structure ξ and an oriented 1–
sphere K ⊂ M embedded transversely to ξ such that the positive orientation of K
coincides with the positive coorientation of ξ. Then in suitable local coordinates
we can identify K with S 1 × {0} ⊂ S 1 × D 2 such that ξ = ker(dθ + r 2 dϕ) and
∂θ corresponds to the positive orientation of K (see Example 2.33). Strictly
speaking, if, as we shall always assume, S 1 is parametrised by 0 ≤ θ ≤ 2π, then
this formula for ξ holds on S 1 × D 2 (δ) for some, possibly small, δ > 0. However,
59

to simplify notation we usually work with S 1 × D 2 as local model.
We say that ξ ′ is obtained from ξ by a Lutz twist along K and write ξ ′ = ξ K
if on S 1 × D 2 the new contact structure ξ ′ is defined by
ξ ′ = ker(h1(r)dθ + h2(r)dϕ)
with (h1(r), h2(r)) as in Figure 14, and ξ ′ coincides with ξ outside S 1 × D 2 .
h2
r = r0
−1 1
Figure 14: Lutz twist.
More precisely, (h1(r), h2(r)) is required to satisfy the conditions
1. h1(r) = −1 and h2(r) = −r 2 near r = 0,
2. h1(r) = 1 and h2(r) = r 2 near r = 1,
3. (h1(r), h2(r)) is never parallel to (h ′ 1 (r), h′ 2 (r)).
This is the same as applying the construction � of Section � 3.2 to the topologically
trivial Dehn surgery given by the matrix
−1
0
0
−1
.
It will be useful later on to understand more precisely the behaviour of the
map fξ ′ : S3 → S 2 . For the definition of this map we assume – this assump-
tion will be justified below – that on S 1 × D 2 the map fξ was defined in terms
60
h1

of the standard metric dθ 2 + du 2 + dv 2 (with u, v cartesian coordinates on D 2
corresponding to the polar coordinates r, ϕ) and the trivialisation ∂θ, ∂u, ∂v of
T(S 1 × D 2 ). Since ξ ′ is spanned by ∂r and h2(r)∂θ − h1(r)∂ϕ (resp. ∂u, ∂v for
r = 0), a vector positively orthogonal to ξ ′ is given by
h1(r)∂θ + h2(r)∂ϕ,
which makes sense even for r = 0. Observe that the ratio h1(r)/h2(r) (for
h2(r) �= 0) is a strictly monotone decreasing function since by the third condition
above we have
(h1/h2) ′ = (h ′ 1h2 − h1h ′ 2)/h 2 2 < 0.
This implies that any value on S 2 other than (1, 0, 0) (corresponding to ∂θ) is
regular for the map fξ ′; the value (1, 0, 0) is attained along the torus {r = r0},
with r0 > 0 determined by h2(r0) = 0, and hence not regular.
If S 1 × D 2 is endowed with the orientation defined by the volume form dθ ∧
r dr ∧ dϕ = dθ ∧ du ∧ dv (so that ξ and ξ ′ are positive contact structures) and
S 2 ⊂ R 3 is given its ‘usual’ orientation defined by the volume form x dy ∧ dz +
y dz ∧ dx + z dx ∧ dy, then
f −1
ξ ′ (−1, 0, 0) = S 1 × {0}
with orientation given by −∂θ, since fξ ′ maps the slices {θ} ×D2 (r0) orientation-
reversingly onto S 2 .
More generally, for any p ∈ S 2 − {(1, 0, 0)} the preimage f −1
ξ ′ (p) (inside the
domain {(θ, r, ϕ): h2(r) < 0} = {r = r0}) is a circle S 1 × {u}, u ∈ D 2 , with
orientation given by −∂θ.
We are now ready to show how to construct a contact structure on M in
any given homotopy class of 2–plane distributions by starting with an arbitrary
contact structure and performing suitable Lutz twists. First we deal with homo-
topy over the 2–skeleton. One way to proceed would be to prove directly, with
notation as above, that d 2 (ξ K , ξ) = −PD[K]. However, it is somewhat easier
to compute d 2 (ξ K , ξ) in the case where ξ is a trivial 2–plane bundle and the
trivialisation of STM is adapted to ξ. Since I would anyway like to present an
alternative argument for computing the effect of a Lutz twist on the Euler class
of the contact structure, and thus relate d 2 (ξ1, ξ2) with the Euler classes of ξ1
and ξ2, it seems opportune to do this first and use it to show the existence of
61

a contact structure with Euler class zero. In the next section we shall actually
discuss a direct geometric proof, due to Gonzalo, of the existence of a contact
structure with Euler class zero.
Recall that the Euler class e(ξ) ∈ H 2 (B; Z) of a 2–plane bundle over a complex
B (of arbitrary dimension) is the obstruction to finding a nowhere zero section
of ξ over the 2–skeleton of B. Since πi(S 1 ) = 0 for i ≥ 2, all higher obstruction
groups H i+1 (B; πi(S 1 )) are trivial, so a 2–dimensional orientable bundle ξ is
trivial if and only if e(ξ) = 0, no matter what the dimension of B.
Now let ξ be an arbitrary cooriented 2–plane distribution on an oriented 3–
manifold M. Then TM ∼ = ξ ⊕ ε 1 , where ε 1 denotes a trivial line bundle. Hence
w2(ξ) = w2(ξ ⊕ ε 1 ) = w2(TM) = 0, and since w2(ξ) is the mod 2 reduction of
e(ξ) we infer that e(ξ) has to be even.
Proposition 3.15. For any even element e ∈ H 2 (M; Z) there is a contact struc-
ture ξ on M with e(ξ) = e.
Proof. Start with an arbitrary contact structure ξ0 on M with e(ξ0) = e0 (which
we know to be even). Given any even e1 ∈ H 2 (M; Z), represent the Poincaré dual
of (e0 − e1)/2 by a collection of embedded oriented circles positively transverse
to ξ0. (Here by (e0 − e1)/2 I mean some class whose double equals e0 − e1; in
the presence of 2–torsion there is of course a choice of such classes.) Choose
a section of ξ0 transverse to the zero section of ξ0, that is, a vector field in
ξ0 with generic zeros. We may assume that there are no zeros on the curves
representing PD −1 (e0 − e1)/2. Now perform a Lutz twist as described above
along these curves and call ξ1 the resulting contact structure. It is easy to see
that in the local model for the Lutz twist a constant vector field tangent to ξ0
along ∂(S 1 ×D 2 (r0)) extends to a vector field tangent to ξ1 over S 1 ×D 2 (r0) with
zeros of index +2 along S 1 × {0} (Figure 15). So the vector field in ξ0 extends
to a vector field in ξ1 with new zeros of index +2 along the curves representing
PD −1 (e1 − e0)/2 (notice that a Lutz twist along a positively transverse knot K
turns K into a negatively transverse knot). Since the self-intersection class of M
in the total space of a vector bundle is Poincaré dual to the Euler class of that
bundle, this proves e(ξ1) = e(ξ0) + e1 − e0 = e1.
We now fix a contact structure ξ0 on M with e(ξ0) = 0 and give M the ori-
entation induced by ξ0 (i.e. the one for which ξ0 is a positive contact structure).
Moreover, we fix a Riemannian metric on M and define X0 as the vector field
62

2
1
2
1
1
1
2
2
1
2 2
Figure 15: Effect of Lutz twist on Euler class.
positively orthonormal to ξ0. Since ξ0 is a trivial plane bundle, X0 extends to an
orthonormal frame X0, X1, X2, hence a trivialisation of STM, with X1, X2 tan-
gent to ξ0 and defining the orientation of ξ0. With these choices, ξ0 corresponds
to the constant map fξ0 : M → (1, 0, 0) ∈ S2 .
Proposition 3.16. Let K ⊂ M be an embedded, oriented circle positively trans-
verse to ξ0. Then d 2 (ξ K 0 , ξ0) = −PD[K].
Proof. Identify a tubular neighbourhood of K ⊂ M with S 1 × D 2 with framing
defined by X1, and ξ0 given in this neighbourhood as the kernel of dθ + r 2 dϕ =
dθ+u dv−v du. We may then change the trivialisation X0, X1, X2 by a homotopy,
fixed outside S 1 × D 2 , such that X0 = ∂θ, X1 = ∂u and X2 = ∂v near K; this
does not change the homotopical data of 2–plane distributions computed via the
Pontrjagin-Thom construction. Then fξ0 is no longer constant, but its image still
does not contain the point (−1, 0, 0).
Now perform a Lutz twist along K × {0}. Our discussion at the beginning
of this section shows that (−1, 0, 0) is a regular value of the map fξ : M → S 2
associated with ξ = ξK −1
0 and fξ (−1, 0, 0) = −K. Hence, by definition of the
obstruction class d2 we have d2 (ξK 0 , ξ0) = −PD[K].
Proof of Theorem 3.1. Let η be a 2–plane distribution on M and ξ0 the contact
structure on M with e(ξ0) = 0 that we fixed earlier on. According to our discus-
sion in Section 3.3.2 and Theorem 2.44, we can find an oriented knot K positively
transverse to ξ0 with −PD[K] = d 2 (η, ξ0). Then d 2 (η, ξ0) = d 2 (ξ K 0 , ξ0) by the
preceding proposition, and therefore d 2 (ξ K 0
63
2
1
, η) = 0.
1
2
1
2
1

We may then assume that η = ξ K 0 on M − D3 , where we choose D 3 so
small that ξ K 0 is in Darboux normal form there (and identical with ξ0). By
Proposition 3.4 we can find a link K ′ in D3 transverse to ξK 0
number l(K ′ ) equal to d3 (η, ξK 0 ).
with self-linking
Now perform a Lutz twist of ξ K 0 along each component of K′ and let ξ be the
resulting contact structure. Since this does not change ξ K 0
of M, we still have d 2 (ξ, η) = 0.
over the 2–skeleton
Observe that f ξ K 0 | D 3 does not contain the point (−1, 0, 0) ∈ S 2 , and – since
f ξ K 0 (D 3 ) is compact – there is a whole neighbourhood U ⊂ S 2 of (−1, 0, 0) not
contained in f ξ K 0 (D 3 ). Let f : S 3 → S 2 be the map used to compute d 3 (ξ, ξ K 0 ),
that is, f coincides on the upper hemisphere with fξ| D 3 and on the lower hemi-
sphere with f ξ K 0 | D 3. By the discussion in Section 3.3, the preimage f −1 (u) of any
u ∈ U − {(−1, 0, 0)} will be a push-off of −K ′ determined by the trivialisation of
ξ K 0 | D 3 = ξ0| D 3. So the linking number of f −1 (u) with f −1 (−1, 0, 0), which is by
definition the Hopf invariant H(f) = d 3 (ξ, ξ K 0 ), will be equal to l(K′ ). By our
choice of K ′ and the additivity of d 3 this implies d 3 (ξ, η) = 0. So ξ is a contact
structure that is homotopic to η as a 2–plane distribution.
3.5 Other existence proofs
Here I briefly summarise the other known existence proofs for contact structures
on 3–manifolds, mostly by pointing to the relevant literature. In spirit, most of
these proofs are similar to the one by Lutz-Martinet: start with a structure theo-
rem for 3–manifolds and show that the topological construction can be performed
compatibly with a contact structure.
3.5.1 Open books
According to a theorem of Alexander [5], cf. [97], every closed, orientable 3–
manifold M admits an open book decomposition. This means that there is
a link L ⊂ M, called the binding, and a fibration f : M − L → S 1 , whose fibres
are called the pages, see Figure 16. It may be assumed that L has a tubular
neighbourhood L × D 2 such that f restricted to L × (D 2 − {0}) is given by
f(θ, r, ϕ) = ϕ, where θ is the coordinate along L and (r, ϕ) are polar coordinates
on D 2 .
At the cost of raising the genus of the pages, one may decrease the number
of components of L, and in particular one may always assume L to be a knot.
64

S 1
f −1 (ϕ)
L
Figure 16: An open book near the binding.
Another way to think of such an open book is as follows. Start with a surface
Σ with boundary ∂Σ = K ∼ = S 1 and a self-diffeomorphism h of Σ with h = id
near K. Form the mapping torus Th = Σh = Σ × [0, 2π]/∼, where ‘∼’ denotes
the identification (p, 2π) ∼ (h(p), 0). Define a 3–manifold M by
M = Th ∪ K×S 1 (K × D 2 ).
This M carries by construction the structure of an open book with binding K
and pages diffeomorphic to Σ.
Here is a slight variation on a simple argument of Thurston and Winkelnkem-
per [101] for producing a contact structure on such an open book (and hence on
any closed, orientable 3–manifold):
Start with a 1–form β0 on Σ with β0 = e t dθ near ∂Σ = K, where θ denotes
the coordinate along K and t is a collar parameter with K = {t = 0} and t < 0 in
the interior of Σ. Then dβ0 integrates to 2π over Σ by Stokes’s theorem. Given
any area form ω on Σ (with total area equal to 2π) satisfying ω = e t dt ∧ dθ
near K, the 2–form ω −dβ0 is, by de Rham’s theorem, an exact 1–form, say dβ1,
where we may assume β1 ≡ 0 near K.
Set β = β0 + β1. Then dβ = ω is an area form (of total area 2π) on Σ and
β = e t dθ near K. The set of 1–forms satisfying these two properties is a convex
set, so we can find a 1–form (still denoted β) on Th which has these properties
when restricted to the fibre over any ϕ ∈ S 1 = [0, 2π]/0∼2π. We may (and shall)
65

equire that β = e t dθ near ∂Th.
Now a contact form α on Th is found by setting α = β+C dϕ for a sufficiently
large constant C ∈ R + , so that in
α ∧ dα = (β + C dϕ) ∧ dβ
the non-zero term dϕ ∧ dβ = dϕ ∧ ω dominates. This contact form can be
extended to all of M by making the ansatz α = h1(r)dθ + h2(r)dϕ on K × D 2 ,
as described in our discussion of the Lutz twist. The boundary conditions in the
present situation are, say,
1. h1(r) = 2 and h2(r) = r 2 near r = 0,
2. h1(r) = e 1−r and h2(r) = C near r = 1.
Observe that subject to these boundary conditions a curve (h1(r), h2(r)) can
be found that does not pass the h2–axis (i.e. with h1(r) never being equal to
zero). In the 3–dimensional setting this is not essential (and the Thurston-
Winkelnkemper ansatz lacked that feature), but it is crucial when one tries to
generalise this construction to higher dimensions. This has recently been worked
out by Giroux and J.-P. Mohsen [57]. This, however, is only the easy part of
their work. Rather strikingly, they have shown that a converse of this result
holds: Given a compact contact manifold of arbitrary dimension, it admits an
open book decomposition that is adapted to the contact structure in the way
described above. Full details have not been published at the time of writing, but
see Giroux’s talk [56] at the ICM 2002.
3.5.2 Branched covers
A theorem of Hilden, Montesinos and Thickstun [63] states that every closed,
orientable 3–manifold M admits a branched covering π: M → S 3 such that
the upstairs branch set is a simple closed curve that bounds an embedded disc.
(Moreover, the cover can be chosen 3–fold and simple, i.e. the monodromy repre-
sentation of π1(S 3 −K), where K is the branching set downstairs (a knot in S 3 ),
represents the meridian of K by a transposition in the symmetric group S3. This,
however, is not relevant for our discussion.)
It follows immediately, as announced in Section 3.3, that every closed, ori-
entable 3–manifold is parallelisable: First of all, S 3 is parallelisable since it car-
ries a Lie group structure (as the unit quaternions, for instance). Given M and
66

a branched covering π: M → S 3 as above, there is a 3–ball D 3 ⊂ M containing
the upstairs branch set. Outside of D 3 , the covering π is unbranched, so the
3–frame on S 3 can be lifted to a frame on M −D 3 . The bundle TM| D 3 is trivial,
so the frame defined along ∂D 3 defines an element of SO(3) (cf. the footnote in
the proof of Theorem 3.10). Since π2(SO(3)) = 0, this frame extends over D 3 .
In [59], Gonzalo uses this theorem to construct a contact structure on every
closed, orientable 3–manifold M, in fact one with zero Euler class: Away from
the branching set one can lift the standard contact structure from S 3 (which
has Euler class zero: a trivialisation is given by two of the three (quaternionic)
Hopf vector fields). A careful analysis of the branched covering map near the
branching set then shows how to extend this contact structure over M (while
keeping it trivial as 2–plane bundle).
A branched covering construction for higher-dimensional contact manifolds is
discussed in [43].
3.5.3 . . . and more
The work of Giroux [52], in which he initiated the study of convex surfaces in
contact 3–manifolds, also contains a proof of Martinet’s theorem.
An entirely different proof, due to S. Altschuler [4], finds contact structures
from solutions to a certain parabolic differential equation for 1–forms on 3–
manifolds. Some of these ideas have entered into the more far-reaching work
of Eliashberg and Thurston on so-called ‘confoliations’ [32], that is, 1–forms sat-
isfying α ∧ dα ≥ 0.
3.6 Tight and overtwisted
The title of this section describes the fundamental dichotomy of contact structures
in dimension 3 that has proved seminal for the development of the field.
In order to motivate the notion of an overtwisted contact structure, as intro-
duced by Eliashberg [21], we consider a ‘full’ Lutz twist as follows. Let (M, ξ) be
a contact 3–manifold and K ⊂ M a knot transverse to ξ. As before, identify K
with S 1 × {0} ⊂ S 1 × D 2 ⊂ M such that ξ = ker(dθ + r 2 dϕ) on S 1 × D 2 . Now
define a new contact structure ξ ′ as in Section 3.4, with (h1(r), h2(r)) now as in
Figure 17, that is, the boundary conditions are now
h1(r) = 1 and h2(r) = r 2 for r ∈ [0, ε] ∪ [1 − ε, 1]
67

for some small ε > 0.
h2
Figure 17: A full Lutz twist.
Lemma 3.17. A full Lutz twist does not change the homotopy class of ξ as a
2–plane field.
Proof. Let (ht 1 (r), ht2 (r)), r, t ∈ [0, 1], be a homotopy of paths such that
1. h 0 1 ≡ 1, h0 2 (r) = r2 ,
2. h 1 i ≡ hi, i = 1, 2,
3. h t i (r) = hi(r) for r ∈ [0, ε] ∪ [1 − ε, 1].
Let χ : [0, 1] → R be a smooth function which is identically zero near r = 0 and
r = 1 and χ(r) > 0 for r ∈ [ε, 1 − ε]. Then
αt = t(1 − t)χ(r)dr + h t 1(r)dθ + h t 2(r)dϕ
is a homotopy from α0 = dθ +r 2 dϕ to α1 = h1(r)dθ +h2(r)dϕ through non-zero
1–forms. This homotopy stays fixed near r = 1, and so it defines a homotopy
between ξ and ξ ′ as 2–plane fields.
68
1
h1

Let r0 be the smaller of the two positive radii with h2(r0) = 0 and consider
the embedding
φ : D 2 (r0) −→ S 1 × D 2
(r, ϕ) ↦−→ (θ(r), r, ϕ),
where θ(r) is a smooth function with θ(r0) = 0, θ(r) > 0 for 0 ≤ r < r0, and
θ ′ (r) = 0 only for r = 0. We may require in addition that θ(r) = θ(0) − r 2 near
r = 0. Then
φ ∗ (h1(r)dθ + h2(r)dϕ) = h1(r)θ ′ (r)dr + h2(r)dϕ
is a differential 1–form on D 2 (r0) which vanishes only for r = 0, and along
∂D 2 (r0) the vector field ∂ϕ tangent to the boundary lies in the kernel of this 1–
form, see Figure 18. In other words, the contact planes ker(h1(r)dθ + h2(r)dϕ)
intersected with the tangent planes to the embedded disc φ(D 2 (r0)) induce a
singular 1–dimensional foliation on this disc with the boundary of this disc as
closed leaf and precisely one singular point in the interior of the disc (Figure 19;
notice that the leaves of this foliation are the integral curves of the vector field
h1(r)θ ′ (r)∂ϕ − h2(r)∂r). Such a disc is called an overtwisted disc.
S 1
r0
Figure 18: An overtwisted disc.
φ(D 2 (r0))
A contact structure ξ on a 3–manifold M is called overtwisted if (M, ξ)
contains an embedded overtwisted disc. In order to justify this terminology,
69
ξ

Figure 19: Characteristic foliation on an overtwisted disc.
observe that in the radially symmetric standard contact structure of Example 2.7,
the angle by which the contact planes turn approaches π/2 asymptotically as r
goes to infinity. By contrast, any contact manifold which has been constructed
using at least one (simple) Lutz twist contains a similar cylindrical region where
the contact planes twist by more than π in radial direction (at the smallest
positive radius r0 with h2(r0) = 0 the twisting angle has reached π).
We have shown the following:
Proposition 3.18. Let ξ be a contact structure on M. By a full Lutz twist along
any transversely embedded circle one obtains an overtwisted contact structure ξ ′
that is homotopic to ξ as a 2–plane distribution. �
Together with the theorem of Lutz and Martinet we find that M contains an
overtwisted contact structure in every homotopy class of 2–plane distributions.
In fact, Eliashberg [21] has proved the following much stronger theorem.
Theorem 3.19 (Eliashberg). On a closed, orientable 3–manifold there is a one-
to-one correspondence between homotopy classes of overtwisted contact structures
and homotopy classes of 2–plane distributions.
This means that two overtwisted contact structures which are homotopic as
2–plane fields are actually homotopic as contact structures and hence isotopic by
Gray’s stability theorem.
Thus, it ‘only’ remains to classify contact structures that are not overtwisted.
In [24] Eliashberg defined tight contact structures on a 3–manifold M as contact
structures ξ for which there is no embedded disc D ⊂ M such that Dξ contains
70

a limit cycle. So, by definition, overtwisted contact structures are not tight. In
that same paper, as mentioned above in Section 2.4.5, Eliashberg goes on to show
the converse with the help of the Elimination Lemma: non-overtwisted contact
structures are tight.
There are various ways to detect whether a contact structure is tight. His-
torically the first proof that a certain contact structure is tight is due to D. Ben-
nequin [9, Cor. 2, p. 150]:
Theorem 3.20 (Bennequin). The standard contact structure ξ0 on S 3 is tight.
The steps of the proof are as follows: (i) First, Bennequin shows that if γ0 is
a transverse knot in (S 3 , ξ0) with Seifert surface Σ, then the self-linking number
of γ satisfies the inequality
l(γ0) ≤ −χ(Σ).
(ii) Second, he introduces an invariant for Legendrian knots; nowadays this
is called the Thurston-Bennequin invariant: Let γ be a Legendrian knot in
(S 3 , ξ0). Take a vector field X along γ transverse to ξ0, and let γ ′ be the push-
off of γ in the direction of X. Then the Thurston-Bennequin invariant tb(γ) is
defined to be the linking number of γ and γ ′ . (This invariant has an extension
to homologically trivial Legendrian knots in arbitrary contact 3–manifolds.)
(iii) By pushing γ in the direction of ±X, one obtains transverse curves γ ±
(either of which is a candidate for γ ′ in (ii)). One of these curves is positively
transverse, the other negatively transverse to ξ0. The self-linking number of γ ± is
related to the Thurston-Bennequin invariant and a further invariant (the rotation
number) of γ. The equation relating these three invariants implies tb(γ) ≤ −χ(Σ),
where Σ again denotes a Seifert surface for γ. In particular, a Legendrian unknot
γ satisfies tb(γ) < 0. This inequality would be violated by the vanishing cycle of
an overtwisted disc (which has tb = 0), which proves that (S 3 , ξ0) is tight.
Remark 3.21. (1) Eliashberg [25] generalised the Bennequin inequality l(γ0) ≤
−χ(Σ) for transverse knots (and the corresponding inequality for the Thurston-
Bennequin invariant of Legendrian knots) to arbitrary tight contact 3–manifolds.
Thus, whereas Bennequin established the tightness (without that name) of the
standard contact structure on S 3 by proving the inequality that bears his name,
that inequality is now seen, conversely, as a consequence of tightness.
(2) In [9] Bennequin denotes the positively (resp. negatively) transverse push-
off of the Legendrian knot γ by γ − (resp. γ + ). This has led to some sign errors in
71

the literature. Notably, the ± in Proposition 2.2.1 of [25], relating the described
invariants of γ and γ ± , needs to be reversed.
Corollary 3.22. The standard contact structure on R 3 is tight.
Proof. This is immediate from Proposition 2.13.
Here are further tests for tightness:
1. A closed contact 3–manifold (M, ξ) is called symplectically fillable if
there exists a compact symplectic manifold (W, ω) bounded by M such that
• the restriction of ω to ξ does not vanish anywhere,
• the orientation of M defined by ξ (i.e. the one for which ξ is positive)
coincides with the orientation of M as boundary of the symplectic manifold
(W, ω) (which is oriented by ω 2 ).
We then have the following result of Eliashberg [20, Thm. 3.2.1], [22] and
Gromov [62, 2.4.D ′ 2 (b)], cf. [10]:
Theorem 3.23 (Eliashberg-Gromov). A symplectically fillable contact structure
is tight.
Example 3.24. The 4–ball D 4 ⊂ R 4 with symplectic form ω = dx1 ∧dy1 +dx2 ∧
dy2 is a symplectic filling of S 3 with its standard contact structure ξ0. This gives
an alternative proof of Bennequin’s theorem.
2. Let ( � M, ˜ ξ) → (M, ξ) be a covering map and contactomorphism. If ( � M, ˜ ξ)
is tight, then so is (M, ξ), for any overtwisted disc in (M, ξ) would lift to an
overtwisted disc in ( � M, ˜ ξ).
Example 3.25. The contact structures ξn, n ∈ N, on the 3–torus T 3 defined by
αn = cos(nθ1)dθ2 + sin(nθ1)dθ3 = 0
are tight: Lift the contact structure ξn to the universal cover R 3 of T 3 ; there the
contact structure is defined by the same equation αn = 0, but now θi ∈ R instead
of θi ∈ R/2πZ ∼ = S 1 . Define a diffeomorphism f of R 3 by
f(x, y, z) = (y/n, z cos y + x sin y, z siny − x cos y) =: (θ1, θ2, θ3).
Then f ∗ αn = dz + x dy, so the lift of ξn to R 3 is contactomorphic to the tight
standard contact structure on R 3 .
72

Notice that it is possible for a tight contact structure to be finitely covered by
an overtwisted contact structure. The first such examples were due to S. Makar-
Limanov [88]. Other examples of this kind have been found by V. Colin [18] and
R. Gompf [58].
3. The following theorem of H. Hofer [65] relates the dynamics of the Reeb
vector field to overtwistedness.
Theorem 3.26 (Hofer). Let α be a contact form on a closed 3–manifold such
that the contact structure kerα is overtwisted. Then the Reeb vector field of α
has at least one contractible periodic orbit.
Example 3.27. The Reeb vector field Rn of the contact form αn of the preceding
example is
Rn = cos(nθ1)∂θ2 + sin(nθ1)∂θ3 .
Thus, the orbits of Rn define constant slope foliations of the 2–tori {θ1 = const.};
in particular, the periodic orbits of Rn are even homologically non-trivial. It
follows, again, that the ξn are tight contact structures on T 3 . (This, admittedly,
amounts to attacking starlings with rice puddings fired from catapults 5 .)
3.7 Classification results
In this section I summarise some of the known classification results for contact
structures on 3–manifolds. By Eliashberg’s Theorem 3.19 it suffices to list the
tight contact structures, up to isotopy or diffeomorphism, on a given closed 3–
manifold.
Theorem 3.28 (Eliashberg [24]). Any tight contact structure on S 3 is isotopic
to the standard contact structure ξ0.
This theorem has a remarkable application in differential topology, viz., it
leads to a new proof of Cerf’s theorem [16] that any diffeomorphism of S 3 ex-
tends to a diffeomorphism of the 4–ball D 4 . The idea is that the above theorem
implies that any diffeomorphism of S 3 is isotopic to a contactomorphism of ξ0.
Eliashberg’s technique [22] of filling by holomorphic discs can then be used to
show that such a contactomorphism extends to a diffeomorphism of D 4 .
5 This turn of phrase originates from [93].
73

As remarked earlier (Remark 2.21), Eliashberg has also classified contact
structures on R 3 . Recall that homotopy classes of 2–plane distributions on S 3
are classified by π3(S 2 ) ∼ = Z. By Theorem 3.19, each of these classes contains
a unique (up to isotopy) overtwisted contact structure. When a point of S 3 is
removed, each of these contact structures induces one on R 3 , and Eliashberg [25]
shows that they remain non-diffeomorphic there. Eliashberg shows further that,
apart from this integer family of overtwisted contact structures, there is a unique
tight contact structure on R 3 (the standard one), and a single overtwisted one
that is ‘overtwisted at infinity’ and cannot be compactified to a contact structure
on S 3 .
In general, the classification of contact structures up to diffeomorphism will
differ from the classification up to isotopy. For instance, on the 3–torus T 3 we
have the following diffeomorphism classification due to Y. Kanda [75]:
Theorem 3.29 (Kanda). Every (positive) tight contact structure on T 3 is con-
tactomorphic to one of the ξn, n ∈ N, described above. For n �= m, the contact
structures ξn and ξm are not contactomorphic.
Giroux [54] had proved earlier that ξn for n ≥ 2 is not contactomorphic to ξ1.
On the other hand, all the ξn are homotopic as 2–plane fields to {dθ1 = 0}.
This shows one way how Eliashberg’s classification Theorem 3.19 for overtwisted
contact structures can fail for tight contact structures:
• There are tight contact structures on T 3 that are homotopic as plane fields
but not contactomorphic.
P. Lisca and G. Matić [82] have found examples of the same kind on homology
spheres by applying Seiberg-Witten theory to Stein fillings of contact manifolds,
cf. also [78].
Eliashberg and L. Polterovich [31] have determined the isotopy classes of
diffeomorphisms of T 3 that contain a contactomorphism of ξ1: they correspond
to exactly those elements of SL(3, Z) = π0(Diff(T 3 )) that stabilise the subspace
0 ⊕ Z 2 corresponding to the coordinates (θ2, θ3). In combination with Kanda’s
result, this allows to give an isotopy classification of tight contact structures
on T 3 . One particular consequence of the result of Eliashberg and Polterovich is
the following:
74

• There are tight contact structures on T 3 that are contactomorphic and
homotopic as plane fields, but not isotopic (i.e. not homotopic through
contact structures).
Again, such examples also exist on homology spheres, as S. Akbulut and
R. Matveyev [2] have shown.
Another aspect of Eliashberg’s classification of overtwisted contact structures
that fails to hold for tight structures is of course the existence of such a structure
in every homotopy class of 2–plane fields, as is already demonstrated by the
classification of contact structures on S 3 . Etnyre and K. Honda [37] have recently
even found an example of a manifold – the connected some of two copies of the
Poincaré sphere with opposite orientations – that does not admit any tight contact
structure at all.
For the classification of tight contact structures on lens spaces and T 2 –bundles
over S 1 see [55], [71] and [72]. A partial classification of tight contact structures
on lens spaces had been obtained earlier in [34].
As further reading on 3–dimensional contact geometry I can recommend the
lucid Bourbaki talk by Giroux [53]. This covers the ground up to Eliashberg’s
classification of overtwisted contact structures and the uniqueness of the tight
contact structure on S 3 .
4 A guide to the literature
In this concluding section I give some recommendations for further reading, con-
centrating on those aspects of contact geometry that have not (or only briefly)
been touched upon in earlier sections.
Two general surveys that emphasise historical matters and describe the de-
velopment of contact geometry from some of its earliest origins are the one by
Lutz [87] and one by the present author [45].
One aspect of contact geometry that I have neglected in these notes is the
Riemannian geometry of contact manifolds (leading, for instance, to Sasakian
geometry). The survey by Lutz has some material on that; D. Blair [11] has
recently published a monograph on the topic.
There have also been various ideas for defining interesting families of contact
structures. Again, the survey by Lutz has something to say on that; one such
75

concept that has exhibited very intriguing ramifications – if this commercial break
be permitted – was introduced in [48].
4.1 Dimension 3
As mentioned earlier, Chapter 8 in [1] is in parts complementary to the present
notes and has some material on surfaces in contact 3–manifolds. Other surveys
and introductory texts on 3–dimensional contact geometry are the introductory
lectures by Etnyre [35] and the Bourbaki talk by Giroux [52]. Good places to
start further reading are two papers by Eliashberg: [24] for the classification of
tight contact structures and [26] for knots in contact 3–manifolds. Concerning
the latter, there is also a chapter by Etnyre [36] in a companion Handbook and an
article by Etnyre and Honda [38] with an extensive introduction to that subject.
The surveys [20] and [27] by Eliashberg are more general in scope, but also
contain material about contact 3–manifolds.
3–dimensional contact topology has now become a fairly wide arena; apart
from the work of Eliashberg, Giroux, Etnyre-Honda and others described earlier,
I should also mention the results of Colin, who has, for instance, shown that
surgery of index one (in particular: taking the connected sum) on a tight contact
3–manifold leads again to a tight contact structure [17].
Finally, Etnyre and L. Ng [40] have compiled a useful list of problems in
3–dimensional contact topology.
4.2 Higher dimensions
The article [46] by the present author contains a survey of what was known at the
time of writing about the existence of contact structures on higher-dimensional
manifolds. One of the most important techniques for constructing contact mani-
folds in higher dimensions is the so-called contact surgery along isotropic spheres
developed by Eliashberg [23] and A. Weinstein [105]. The latter is a very readable
paper. For a recent application of this technique see [49]. Other constructions
of contact manifolds (branched covers, gluing along codimension 2 contact sub-
manifolds) are described in my paper [43].
Odd-dimensional tori are of course amongst the manifolds with the simplest
global description, but they do not easily lend themselves to the construction of
contact structures. In [86] Lutz found a contact structure on T 5 ; since then it has
been one of the prize questions in contact geometry to find a contact structure on
76

higher-dimensional tori. That prize, as it were, recently went to F. Bourgeois [13],
who showed that indeed all odd-dimensional tori do admit a contact structure.
His construction uses the result of Giroux and Mohsen [56], [57] about open book
decompositions adapted to contact structures in conjunction with the original
proof of Lutz. With the help of the branched cover theorem described in [43] one
can conclude further that every manifold of the form M × Σ with M a contact
manifold and Σ a surface of genus at least 1 admits a contact structure.
Concerning the classification of contact structures in higher dimensions, the
first steps have been taken by Eliashberg [28] with the development of con-
tact homology, which has been taken further in [29]. This has been used by
I. Ustilovsky [102] to show that on S 4n+1 there exist infinitely many non-isomor-
phic contact structures that are homotopically equivalent (in the sense that they
induce the same almost contact structure, i.e. reduction of the structure group
of TS 4n+1 to 1 × U(2n)). Earlier results in this direction can be found in [44] in
the context of various applications of contact surgery.
4.3 Symplectic fillings
A survey on the various types of symplectic fillings of contact manifolds is given
by Etnyre [33], cf. also the survey by Bennequin [10]. Etnyre and Honda [39]
have recently shown that certain Seifert fibred 3–manifolds M admit tight con-
tact structures ξ that are not symplectically semi-fillable, i.e. there is no sym-
plectic filling W of (M, ξ) even if W is allowed to have other contact boundary
components. That paper also contains a good update on the general question of
symplectic fillability.
A related question is whether symplectic manifolds can have disconnected
boundary of contact type (this corresponds to a stronger notion of symplectic
filling defined via a Liouville vector field transverse to the boundary and pointing
outwards). For (boundary) dimension 3 this is discussed by D. McDuff [91];
higher-dimensional symplectic manifolds with disconnected boundary of contact
type have been constructed in [42].
4.4 Dynamics of the Reeb vector field
In a seminal paper, Hofer [65] applied the method of pseudo-holomorphic curves,
which had been introduced to symplectic geometry by Gromov [62], to solve
(for large classes of contact 3–manifolds) the so-called Weinstein conjecture [104]
77

concerning the existence of periodic orbits of the Reeb vector field of a given
contact form. (In fact, one of the remarkable aspects of Hofer’s work is that in
many instances it shows the existence of a periodic orbit of the Reeb vector field
of any contact form defining a given contact structure.) A Bourbaki talk on the
state of the art around the time when Weinstein formulated the conjecture that
bears his name was given by N. Desolneux-Moulis [19]; another Bourbaki talk by
F. Laudenbach describes Hofer’s contribution to the problem.
The textbook by Hofer and E. Zehnder [70] addresses these issues, although its
main emphasis, as is clear from the title, lies more in the direction of symplectic
geometry and Hamiltonian dynamics. Two surveys by Hofer [66], [67], and one
by Hofer and M. Kriener [68], are more directly concerned with contact geometry.
Of the three, [66] may be the best place to start, since it derives from a course of
five lectures. In collaboration with K. Wysocki and Zehnder, Hofer has expanded
his initial ideas into a far-reaching project on the characterisation of contact
manifolds via the dynamics of the Reeb vector field, see e.g. [69].
References
[1] B. Aebischer, M. Borer, M. Kälin, Ch. Leuenberger and H.M. Reimann,
Symplectic Geometry, Progr. Math. 124, Birkhäuser, Basel (1994).
[2] S. Akbulut and R. Matveyev, A note on contact structures, Pacific J. Math.
182 (1998), 201–204.
[3] D.N. Akhiezer, Homogeneous complex manifolds, in: Several Complex Vari-
ables IV (S.G. Gindikin, G.M. Khenkin, eds.), Encyclopaedia Math. Sci. 10,
Springer, Berlin (1990), 195–244.
[4] S. Altschuler, A geometric heat flow for one-forms on three-dimensional
manifolds, Illinois J. Math. 39 (1995), 98–118.
[5] J.W. Alexander, A lemma on systems of knotted curves, Proc. Nat. Acad.
Sci. U.S.A. 9 (1923), 93–95.
[6] V.I. Arnold, Characteristic class entering in quantization conditions, Funct.
Anal. Appl. 1 (1967), 1–13.
[7] V.I. Arnold, Mathematical Methods of Classical Mechanics, Grad. Texts in
Math. 60, Springer, Berlin (1978).
78

[8] A. Banyaga, The Structure of Classical Diffeomorphism Groups, Math. Appl.
400, Kluwer, Dordrecht (1997).
[9] D. Bennequin, Entrelacements et équations de Pfaff, in: IIIe Rencontre de
Géométrie du Schnepfenried, vol. 1, Astérisque 107–108 (1983), 87–161.
[10] D. Bennequin, Topologie symplectique, convexité holomorphe et structures
de contact (d’après Y. Eliashberg, D. McDuff et al.), in: Séminaire Bourbaki,
vol. 1989/90, Astérisque 189–190 (1990), 285–323.
[11] D.E. Blair, Riemannian Geometry of Contact and Symplectic Manifolds,
Progr. Math. 203, Birkhäuser, Basel (2002).
[12] W.M. Boothby, Transitivity of the automorphisms of certain geometric struc-
tures, Trans. Amer. Math. Soc. 137 (1969), 93–100.
[13] F. Bourgeois, Odd dimensional tori are contact manifolds, Int. Math. Res.
Notices 2002, 1571–1574.
[14] G.E. Bredon, Topology and Geometry, Grad. Texts in Math. 139, Springer,
Berlin (1993).
[15] Th. Bröcker and K. Jänich, Einführung in die Differentialtopologie, Springer,
Berlin (1973).
[16] J. Cerf, Sur les difféomorphismes de la sphère de dimension trois (Γ4 = 0),
Lecture Notes in Math. 53, Springer, Berlin (1968).
[17] V. Colin, Chirurgies d’indice un et isotopies de sphères dans les variétés de
contact tendues, C. R. Acad. Sci. Paris Sér. I Math. 324 (1997), 659–663.
[18] V. Colin, Recollement de variétés de contact tendues, Bull. Soc. Math. France
127 (1999), 43–69.
[19] N. Desolneux-Moulis, Orbites périodiques des systèmes hamiltoniens au-
tonomes (d’après Clarke, Ekeland-Lasry, Moser, Rabinowitz, Weinstein), in:
Séminaire Bourbaki, vol. 1979/80, Lecture Notes in Math. 842, Springer,
Berlin (1981), 156–173.
79

[20] Y. Eliashberg, Three lectures on symplectic topology in Cala Gonone: Basic
notions, problems and some methods, in: Conference on Differential Geo-
metry and Topology (Sardinia, 1988), Rend. Sem. Fac. Sci. Univ. Cagliari
58 (1988), suppl., 27–49.
[21] Y. Eliashberg, Classification of overtwisted contact structures on 3–mani-
folds, Invent. Math. 98 (1989), 623–637.
[22] Y. Eliashberg, Filling by holomorphic discs and its applications, in: Geo-
metry of Low-Dimensional Manifolds (Durham, 1989) vol. 2, London Math.
Soc. Lecture Note Ser. 151, Cambridge University Press (1990), 45–67.
[23] Y. Eliashberg, Topological characterization of Stein manifolds of dimension
> 2, Internat. J. Math. 1 (1990), 29–46.
[24] Y. Eliashberg, Contact 3–manifolds twenty years since J. Martinet’s work,
Ann. Inst. Fourier (Grenoble) 42 (1992), no. 1-2, 165–192.
[25] Y. Eliashberg, Classification of contact structures on R 3 , Internat. Math.
Res. Notices 1993, 87–91.
[26] Y. Eliashberg, Legendrian and transversal knots in tight contact 3–mani-
folds, in: Topological Methods in Modern Mathematics (Stony Brook, 1991),
Publish or Perish, Houston (1993), 171–193.
[27] Y. Eliashberg, Symplectic topology in the nineties, Differential Geom. Appl.
9 (1998), 59–88.
[28] Y. Eliashberg, Invariants in contact topology, in: Proceedings of the Interna-
tional Congress of Mathematicians (Berlin, 1998) vol. II, Doc. Math. 1998,
Extra Vol. II, 327–338.
[29] Y. Eliashberg, A. Givental and H. Hofer, Introduction to symplectic field
theory, Geom. Funct. Anal. 2000, Special Volume, Part II, 560–673.
[30] Y. Eliashberg and N. Mishachev, Introduction to the h-principle, Grad. Stud.
Math. 48, American Mathematical Society, Providence (2002).
[31] Y. Eliashberg and L. Polterovich, New applications of Luttinger’s surgery,
Comment. Math. Helv. 69 (1994), 512–522.
80

[32] Y. Eliashberg and W.P. Thurston, Confoliations, Univ. Lecture Ser. 13,
American Mathematical Society, Providence (1998).
[33] J.B. Etnyre, Symplectic convexity in low-dimensional topology, Topology
Appl. 88 (1998), 3–25.
[34] J.B. Etnyre, Tight contact structures on lens spaces, Commun. Contemp.
Math. 2 (2000), 559–577; erratum: ibid. 3 (2001), 649–652.
[35] J.B. Etnyre, Introductory lectures on contact geometry, in: Topology and
Geometry of Manifolds (Athens, GA, 2001), Proc. Sympos. Pure Math. 71,
American Mathematical Society, Providence (2003), 81–107.
[36] J.B. Etnyre, Legendrian and transverse knots, Handbook of Knot Theory
(W. Menasco, M. Thistlethwaite, eds.), to appear.
[37] J.B. Etnyre and K. Honda, On the nonexistence of tight contact structures,
Ann. of Math. (2) 153 (2001), 749–766.
[38] J.B. Etnyre and K. Honda, Knots and contact geometry I: Torus knots and
the figure eight knot, J. Symplectic Geom. 1 (2001), 63–120.
[39] J.B. Etnyre and K. Honda, Tight contact structures with no symplectic
fillings, Invent. Math. 148 (2002), 609–626.
[40] J.B. Etnyre and L.L. Ng, Problems in low dimensional contact topology, in:
Topology and Geometry of Manifolds (Athens, GA, 2001), Proc. Sympos.
Pure Math. 71, American Mathematical Society, Providence (2003), 337–
357.
[41] H. Freudenthal, Zum Hopfschen Umkehrhomomorphismus, Ann. of Math.
(2) 38 (1937), 847–853.
[42] H. Geiges, Symplectic manifolds with disconnected boundary of contact type,
Internat. Math. Res. Notices 1994, 23–30.
[43] H. Geiges, Constructions of contact manifolds, Math. Proc. Cambridge Phi-
los. Soc. 121 (1997), 455–464.
[44] H. Geiges, Applications of contact surgery, Topology 36 (1997), 1193–1220.
81

[45] H. Geiges, A brief history of contact geometry and topology, Expo. Math.
19 (2001), 25–53.
[46] H. Geiges, Contact topology in dimension greater than three, in: European
Congress of Mathematics (Barcelona, 2000) vol. 2, Progress in Math. 202,
Birkhäuser, Basel (2001), 535–545.
[47] H. Geiges, h-Principles and flexibility in geometry, Mem. Amer. Math. Soc.
164 (2003), no. 779.
[48] H. Geiges and J. Gonzalo, Contact geometry and complex surfaces, Invent.
Math. 121 (1995), 147–209.
[49] H. Geiges and C.B. Thomas, Contact structures, equivariant spin bordism,
and periodic fundamental groups, Math. Ann. 320 (2001), 685–708.
[50] V.L. Ginzburg, On closed characteristics of 2–forms, Ph.D. Thesis, Berkeley
(1990).
[51] V.L. Ginzburg, Calculation of contact and symplectic cobordism groups,
Topology 31 (1992), 767–773.
[52] E. Giroux, Convexité en topologie de contact, Comment. Math. Helv. 66
(1991), 637–677.
[53] E. Giroux, Topologie de contact en dimension 3 (autour des travaux de Yakov
Eliashberg), in: Séminaire Bourbaki, vol. 1992/93, Astérisque 216 (1993),
7–33.
[54] E. Giroux, Une structure de contact, même tendue, est plus ou moins tordue,
Ann. Sci. École Norm. Sup. (4) 27 (1994), 697–705.
[55] E. Giroux, Structures de contact en dimension trois et bifurcations des feuil-
letages de surfaces, Invent. Math. 141 (2000), 615–689.
[56] E. Giroux, Géométrie de contact: de la dimension trois vers les dimensions
supérieures, in: Proceedings of the International Congress of Mathematicians
(Beijing, 2002) vol. II, Higher Education Press, Beijing (2002), 405–414.
[57] E. Giroux and J.-P. Mohsen, Structures de contact et fibrations symplec-
tiques au-dessus du cercle, in preparation.
82

[58] R.E. Gompf, Handlebody construction of Stein surfaces, Ann. of Math. (2)
148 (1998), 619–693.
[59] J. Gonzalo, Branched covers and contact structures, Proc. Amer. Math. Soc.
101 (1987), 347–352.
[60] D.H. Gottlieb, Partial transfers, in: Geometric Applications of Homotopy
Theory I, Lecture Notes in Math. 657, Springer, Berlin (1978), 255–266.
[61] J.W. Gray, Some global properties of contact structures, Ann. of Math. (2)
69 (1959), 421–450.
[62] M. Gromov, Pseudoholomorphic curves in symplectic manifolds, Invent.
Math. 82 (1985), 307–347.
[63] H.M. Hilden, J.M. Montesinos and T. Thickstun, Closed oriented 3–
manifolds as 3–fold branched coverings of S 3 of special type, Pacific J. Math.
65 (1976), 65–76.
[64] M.W. Hirsch, Differential Topology, Grad. Texts in Math. 33, Springer,
Berlin (1976).
[65] H. Hofer, Pseudoholomorphic curves in symplectizations with applications to
the Weinstein conjecture in dimension 3, Invent. Math. 114 (1993), 515–563.
[66] H. Hofer, Holomorphic curves and dynamics in dimension three, in: Sym-
plectic Geometry and Topology (Park City, 1997), IAS/Park City Math. Ser.
7, American Mathematical Society, Providence (1999), 35–101.
[67] H. Hofer, Holomorphic curves and real three-dimensional dynamics, Geom.
Funct. Anal. 2000, Special Volume, Part II, 674–704.
[68] H. Hofer and M. Kriener, Holomorphic curves in contact dynamics, in: Dif-
ferential Equations (La Pietra, 1996), Proc. Sympos. Pure Math. 65, Amer-
ican Mathematical Society, Providence (1999), 77–131.
[69] H. Hofer, K. Wysocki and E. Zehnder, A characterisation of the tight three-
sphere, Duke Math. J. 81 (1995), 159–226; erratum: ibid. 89 (1997), 603–
617.
[70] H. Hofer and E. Zehnder, Symplectic Invariants and Hamiltonian Dynamics
Birkhäuser, Basel (1994).
83

[71] K. Honda, On the classification of tight contact structures I, Geom. Topol.
4 (2000), 309–368.
[72] K. Honda, On the classification of tight contact structures II, J. Differential
Geom. 55 (2000), 83–143.
[73] H. Hopf, Zur Algebra der Abbildungen von Mannigfaltigkeiten, J. Reine
Angew. Math. 105 (1930), 71–88.
[74] D. Husemoller, Fibre Bundles (3rd edition), Grad. Texts in Math. 20,
Springer, Berlin (1994).
[75] Y. Kanda, The classification of tight contact structures on the 3–torus,
Comm. Anal. Geom. 5 (1997), 413–438.
[76] R.C. Kirby, The Topology of 4–Manifolds, Lecture Notes in Math. 1374,
Springer, Berlin (1989).
[77] A.A. Kosinski, Differential Manifolds, Academic Press, Boston (1993).
[78] P.B. Kronheimer and T.S. Mrowka, Monopoles and contact structures, In-
vent. Math. 130 (1997), 209–255.
[79] F. Laudenbach, Orbites périodiques et courbes pseudo-holomorphes, appli-
cation à la conjecture de Weinstein en dimension 3 (d’après H. Hofer et al.),
in: Séminaire Bourbaki, vol. 1993/1994, Astérisque 227 (1995), 309–333.
[80] P. Libermann and C.-M. Marle, Symplectic Geometry and Analytical Me-
chanics, Math. Appl. 35, Reidel, Dordrecht (1987).
[81] W.B.R. Lickorish, A representation of orientable combinatorial 3–manifolds,
Ann. of Math. (2) 76 (1962), 531–540.
[82] P. Lisca and G. Matić, Tight contact structures and Seiberg-Witten invari-
ants, Invent. Math. 129 (1997), 509–525.
[83] R. Lutz, Sur l’existence de certaines formes différentielles remarquables sur
la sphère S 3 , C. R. Acad. Sci. Paris Sér. A–B 270 (1970), A1597–A1599.
[84] R. Lutz, Sur quelques propriétés des formes differentielles en dimension trois,
Thèse, Strasbourg (1971).
84

[85] R. Lutz, Structures de contact sur les fibrés principaux en cercles de dimen-
sion trois, Ann. Inst. Fourier (Grenoble), 27 (1977), no. 3, 1–15.
[86] R. Lutz, Sur la géométrie des structures de contact invariantes, Ann. Inst.
Fourier (Grenoble) 29 (1979), no. 1, 283–306.
[87] R. Lutz, Quelques remarques historiques et prospectives sur la géométrie de
contact, in: Conference on Differential Geometry and Topology (Sardinia,
1988), Rend. Sem. Fac. Sci. Univ. Cagliari 58 (1988), suppl., 361–393.
[88] S. Makar-Limanov, Tight contact structures on solid tori, Trans. Amer.
Math. Soc. 350 (1998), 1013–1044.
[89] J. Martinet, Sur les singularités des formes différentielles, Ann Inst. Fourier
(Grenoble) 20 (1970) no. 1, 95–178.
[90] J. Martinet, Formes de contact sur les variétés de dimension 3, in: Proc.
Liverpool Singularities Sympos. II, Lecture Notes in Math. 209, Springer,
Berlin (1971), 142–163.
[91] D. McDuff, Symplectic manifolds with contact type boundaries, Invent.
Math. 103 (1991), 651–671.
[92] D. McDuff and D. Salamon, Introduction to Symplectic Topology (2nd edi-
tion), Oxford University Press, 1998.
[93] S. Milligan, The Starlings, in: The Goon Show, Series 4, Special Programme,
broadcast 31st August 1954.
[94] J.W. Milnor, Topology from the Differentiable Viewpoint, The University
Press of Virginia, Charlottesville (1965).
[95] J.W. Milnor and J.D. Stasheff, Characteristic Classes, Ann. of Math. Studies
76, Princeton University Press, Princeton (1974).
[96] J. Moser, On the volume elements on a manifold, Trans. Amer. Math. Soc.
120 (1965), 286–294.
[97] D. Rolfsen, Knots and Links, Publish or Perish, Houston (1976).
[98] N. Saveliev, Lectures on the Topology of 3–Manifolds, de Gruyter, Berlin
(1999).
85

[99] N. Steenrod, The Topology of Fibre Bundles, Princeton University Press,
Princeton (1951).
[100] I. Tamura, Topology of Foliations: An Introduction, Transl. Math. Monogr.
97, American Mathematical Society, Providence, 1992.
[101] W.P. Thurston and H.E. Winkelnkemper, On the existence of contact forms,
Proc. Amer. Math. Soc. 52 (1975), 345–347.
[102] I. Ustilovsky, Infinitely many contact structures on S 4m+1 , Internat. Math.
Res. Notices 1999, 781–791.
[103] A.H. Wallace, Modifications and cobounding manifolds, Canad. J. Math.
12 (1960), 503–528.
[104] A. Weinstein, On the hypotheses of Rabinowitz’ periodic orbit theorem, J.
Differential Equations 33 (1979), 353–358.
[105] A. Weinstein, Contact surgery and symplectic handlebodies, Hokkaido
Math. J. 20 (1991), 241–251.
86